Lithium battery data anomaly detection method and system based on machine learning model

Abstract

The invention belongs to the technical field of lithium battery detection, and discloses a lithium battery data anomaly detection method and system based on a machine learning model, initial battery data is processed to obtain target battery data convenient for subsequent model processing and analysis, and according to a data distribution diagram corresponding to each type of feature data, the target battery data can be analyzed according to the data distribution diagram corresponding to each type of feature data. The corresponding coefficient factor is determined to process the target battery data, so that the adaptability and the flexibility of battery detection can be guaranteed, and the normal overall distribution of the processed target battery data can be effectively guaranteed; then, the processed target battery data can be predicted based on at least two machine learning models, corresponding abnormal predicted values are obtained, the sensitivity to noise data and abnormal values is reduced through a hybrid model prediction mode, and therefore the overall robustness and generalization ability are improved. And the effectiveness and the accuracy of battery anomaly detection are ensured.

Description

Technical Field

[0001] The present application belongs to the field of lithium battery detection technology, and in particular relates to a lithium battery data anomaly detection method and system based on a machine learning model. Background Art

[0002] With the widespread use of lithium batteries in electric vehicles, mobile devices and energy storage systems, their safety and reliability have received great attention. The performance of lithium batteries will degrade with the increase of usage time. This degradation may lead to safety issues such as decreased battery capacity, increased internal resistance and even thermal runaway. Therefore, real-time monitoring of the health status of lithium batteries, especially the detection of abnormal data points, is crucial.

[0003] Traditional methods of detecting abnormal data points are mainly based on statistics. For example, they rely on the statistical characteristics of data (mean or standard deviation, etc.) to identify anomalies. However, this method usually requires assumptions about data distribution and cannot guarantee the effectiveness and accuracy of data detection. Secondly, it has low processing efficiency for high-dimensional data, and cannot guarantee the adaptability and flexibility of detection when facing dynamically changing data or newly emerging abnormal data types, which in turn affects the overall anomaly detection effect. Summary of the invention

[0004] In order to solve the above-mentioned technical defects that the traditional abnormal data point detection method needs to make assumptions about data distribution, cannot guarantee the effectiveness and accuracy of data detection, and secondly has low processing efficiency for high-dimensional data, and cannot guarantee the adaptability and flexibility of detection when facing dynamically changing data or newly emerging abnormal data types, thus affecting the overall abnormal detection effect, this application proposes a lithium battery data anomaly detection method and system based on a machine learning model, and its technical solution is as follows: In a first aspect, an embodiment of the present application provides a method for detecting anomaly in lithium battery data based on a machine learning model, comprising: Processing at least two groups of initial battery data to obtain target battery data; wherein each group of initial battery data includes at least two types of characteristic data; According to the data distribution diagram corresponding to each type of characteristic data, the corresponding coefficient factors are determined, and the target battery data is processed based on all the coefficient factors; Based on at least two machine learning models, the processed target battery data is predicted respectively to obtain corresponding abnormal prediction values, and the battery abnormality detection result is obtained according to all the abnormal prediction values.

[0005] In an optional solution of the first aspect, processing at least two sets of initial battery data to obtain target battery data includes: Based on a preset conversion function, at least two groups of initial battery data are converted respectively, and all the converted initial battery data are merged to obtain merged battery data; When missing data is detected in the merged battery data, supplementary data is obtained according to the characteristic data corresponding to the type of the missing data, and the supplementary data is filled into the merged battery data; or When missing data is detected in the merged battery data, determining preset data corresponding to the type of the missing data, and filling the preset data into the merged battery data; In the combined battery data after the filling process, statistical processing is performed on each characteristic data, and all statistical results are integrated into the combined battery data after the filling process to obtain the target battery data.

[0006] In yet another optional solution of the first aspect, determining corresponding coefficient factors according to a data distribution graph corresponding to each type of characteristic data includes: According to the data distribution graph corresponding to each type of feature data, the corresponding tail distribution probability is calculated; Based on the probability interval in which each tail distribution probability lies, the corresponding coefficient factor is queried in a preset probability interval-coefficient factor list; wherein the preset probability interval-coefficient factor list includes at least two groups of standard probability intervals and a standard coefficient factor corresponding to each group of standard probability intervals.

[0007] In another optional solution of the first aspect, the target battery data is processed based on all coefficient factors, including: According to the preset quantile coefficient, the quantile corresponding to each type of characteristic data is determined, and based on the quantile corresponding to each type of characteristic data and the coefficient factor, the corresponding quantile interval is calculated; When it is detected in each type of feature data that any feature data is not within the corresponding quantile interval, the feature data is eliminated.

[0008] In another optional solution of the first aspect, after processing the target battery data based on all coefficient factors and before predicting the processed target battery data based on at least two machine learning models respectively, the method further includes: According to the data heat map corresponding to all types of feature data, determine the correlation coefficient between any two types of feature data; When the correlation coefficient between any two types of feature data is in the strong correlation coefficient interval, the corresponding two types of feature data are screened; When the correlation coefficient between any two types of feature data is in a weak correlation coefficient interval, the corresponding two types of feature data are retained.

[0009] In another optional solution of the first aspect, obtaining a battery abnormality detection result according to all abnormality prediction values ​​includes: When the number of abnormal prediction values ​​detected in all abnormal prediction values ​​that are within a preset abnormal value interval exceeds a preset number threshold, it is determined that the battery abnormality detection result is a battery data abnormality; When the number of abnormal prediction values ​​detected in all abnormal prediction values ​​that are within a preset abnormal value interval does not exceed a preset number threshold, it is determined that the battery abnormality detection result is that the battery data is normal.

[0010] In another optional solution of the first aspect, after obtaining the battery abnormality detection result according to all abnormality prediction values, the method further includes: When the battery abnormality detection result is that the battery data is abnormal, the processed target battery data is predicted based on at least two classification learning models to obtain corresponding classification prediction values; All classification prediction values ​​are concatenated to obtain a classification feature matrix, and the classification feature matrix is ​​input into the logistic regression model to obtain the abnormal classification result; Update the battery anomaly detection result to the anomaly classification result.

[0011] In a second aspect, an embodiment of the present application provides a battery data anomaly detection system based on a machine learning model, comprising: A first processing module, used to process at least two groups of initial battery data to obtain target battery data; wherein each group of initial battery data includes at least two types of feature data; A second processing module, used to determine corresponding coefficient factors according to a data distribution diagram corresponding to each type of characteristic data, and process the target battery data based on all coefficient factors; The result prediction module is used to predict the processed target battery data based on at least two machine learning models, obtain corresponding abnormal prediction values, and obtain battery abnormality detection results based on all abnormal prediction values.

[0012] In a third aspect, an embodiment of the present application further provides a battery data anomaly detection system based on a machine learning model, including a processor and a memory; The processor is connected to the memory; A memory for storing executable program codes; The processor runs the program corresponding to the executable program code by reading the executable program code stored in the memory, so as to implement the battery data anomaly detection method based on the machine learning model provided by the first aspect of the embodiment of the present application or any one of the implementation methods of the first aspect.

[0013] In a fourth aspect, an embodiment of the present application provides a computer storage medium, which stores a computer program. The computer program includes program instructions. When the program instructions are executed by a processor, the battery data anomaly detection method based on a machine learning model provided by the first aspect of the embodiment of the present application or any one of the implementation methods of the first aspect can be implemented.

[0014] Beneficial effects of this application: When performing abnormal detection of battery data, at least two groups of initial battery data are processed to obtain target battery data; the corresponding coefficient factors are determined according to the data distribution diagram corresponding to each type of characteristic data, and the target battery data is processed based on all the coefficient factors; the processed target battery data is predicted based on at least two machine learning models to obtain corresponding abnormal prediction values, and the battery abnormality detection results are obtained based on all abnormal prediction values. By processing the initial battery data, target battery data that is convenient for subsequent model processing and analysis is obtained, and the corresponding coefficient factors can be determined according to the data distribution diagram corresponding to each type of characteristic data to process the target battery data, which can not only ensure the adaptability and flexibility of battery detection, but also effectively ensure that the overall distribution of the processed target battery data is normal; then, the processed target battery data can be predicted based on at least two machine learning models to obtain corresponding abnormal prediction values, so as to reduce the sensitivity to noise data and abnormal values ​​through hybrid model prediction, and the diversity of the model can be used to improve the overall adaptability, robustness and generalization ability when the battery data distribution changes, thereby ensuring the effectiveness and accuracy of battery abnormality detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 An overall flow chart of a battery data anomaly detection method based on a machine learning model provided in an embodiment of the present application; Figure 2 A schematic diagram of the effect of processing target battery data based on all coefficient factors provided in an embodiment of the present application; Figure 3 A schematic diagram of the prediction effect of a machine learning model provided in an embodiment of the present application; Figure 4 A schematic diagram of the prediction effect of another machine learning model provided in an embodiment of the present application; Figure 5 A structural diagram of a battery data anomaly detection system based on a machine learning model provided in an embodiment of the present application; Figure 6 A structural schematic diagram of another battery data anomaly detection system based on a machine learning model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0018] In the following introduction, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance. The following introduction provides multiple embodiments of the present application, and different embodiments can be replaced or combined, so the present application can also be considered to include all possible combinations of the same and / or different embodiments recorded. Therefore, if one embodiment includes features A, B, and C, and another embodiment includes features B and D, then the present application should also be considered to include embodiments containing one or more of all other possible combinations of A, B, C, and D, although the embodiment may not be clearly recorded in the following text.

[0019] The following description provides examples and does not limit the scope, applicability or examples set forth in the claims. Changes may be made to the functions and arrangements of the elements described without departing from the scope of the present application. Various processes or components may be appropriately omitted, substituted or added to each example. For example, the described method may be performed in an order different from the order described, and various steps may be added, omitted or combined. In addition, features described in some examples may be combined in other examples.

[0020] It should be emphasized that the traditional method of detecting abnormal data points in batteries is mainly based on statistics, such as but not limited to relying on the statistical characteristics of the data (mean, standard deviation or variance, etc.) to identify abnormal data points, but there are certain disadvantages: First, it is necessary to make certain assumptions about the distribution of battery data. If the assumptions do not conform to the true distribution of the actual data, the accuracy and effectiveness of the detection results of the abnormal data points will be seriously affected; second, when processing high-dimensional data, as the data dimension increases, the data will become increasingly sparse, making it more difficult to detect abnormal data points; third, when faced with dynamically changing data or newly emerging abnormal data types, the adaptability and flexibility of the detection cannot be guaranteed, and the detection effect on noise data points is too sensitive, which in turn affects the overall abnormality detection effect.

[0021] Secondly, the traditional method of detecting abnormal battery data points relies on a single machine learning algorithm, but it has the following defects: first, a single machine learning algorithm may perform well on certain types of data sets, but not well on other types of data sets, which can easily limit the generalization ability of the algorithm; second, the single machine learning algorithm is not robust enough to noise data and abnormal data points, that is, it is easily affected by noise data or abnormal data points, resulting in poor overall anomaly detection effect; third, the single machine learning algorithm has strong feature dependence, that is, it is easy to affect the overall anomaly detection effect due to improper feature selection, and it is also easy to reach bottlenecks in specific performance indicators, and thus cannot guarantee the accuracy of the detection results and the false alarm rate.

[0022] Based on this, this application will explain the battery data anomaly detection method based on a machine learning model in combination with one or more embodiments proposed below.

[0023] See next Figure 1 , Figure 1 An overall flow chart of a battery data anomaly detection method based on a machine learning model provided in an embodiment of the present application is shown.

[0024] like Figure 1 As shown, the battery data anomaly detection method based on the machine learning model may at least include the following steps: Step 102: Process at least two groups of initial battery data to obtain target battery data.

[0025] In the embodiment of the present application, the battery data anomaly detection method based on the machine learning model can be applied to, but not limited to, a test terminal, which can establish a connection with a battery test cabinet to analyze and process multiple groups of initial battery data recorded by the battery test cabinet to obtain corresponding anomaly detection results. Wherein, each group of initial battery data can be understood as at least two types of characteristic data recorded by the battery test cabinet during the battery test process, such as but not limited to multiple operating temperatures, multiple charging times, multiple charging cycles, multiple charging currents, multiple charging voltages, and multiple battery capacities of the battery in the charging state, each of which can have corresponding charging time, charging cycle number, charging current, charging voltage, and battery capacity; it can also include multiple operating temperatures, multiple discharge times, multiple discharge cycles, multiple load currents, multiple load voltages, and multiple battery capacities of the battery in the discharging state, each of which can have corresponding discharge time, discharge cycle number, load current, load voltage, and battery capacity; it can also include multiple branch currents, multiple branch voltages, multiple battery impedances, and multiple estimated charging transfer resistances of the battery in the impedance state, and is not limited to the data types mentioned above.

[0026] Here, when the test terminal analyzes and processes multiple groups of initial battery data, it can also apply multiple unsupervised learning machine learning models to respectively predict abnormalities of the processed initial battery data, so as to obtain the abnormal prediction value corresponding to each machine learning model, and obtain the final battery abnormality detection result according to all abnormal prediction values. In the embodiment of the present application, the types of multiple machine learning models can include but are not limited to density-based spatial clustering learning model (Density-Based Spatial Clustering of Applications with Noise, DBSCAN), isolation forest learning model (Isolation Forest) and density-based anomaly detection learning model (Local Outlier Factor, LOF). The density-based spatial clustering learning model can be good at processing outliers in different density areas, the isolation forest learning model can ensure the efficiency of data processing, and the density-based anomaly detection learning model can quantify the degree of abnormality of each sample point, which is suitable for data with different density dispersion conditions of different clusters, that is, different types of machine learning models can reduce the sensitivity to noise data and outliers, and can also improve the overall robustness and generalization ability.

[0027] In addition, after obtaining the battery abnormality detection result, when determining that the battery data is abnormal, the test terminal can use multiple classification learning models to classify and predict the processed initial battery data respectively, so as to obtain the classification prediction value corresponding to each classification learning model, and the final abnormal classification result can be obtained according to all the classification prediction values. In the embodiment of the present application, the types of multiple classification learning models can include but are not limited to decision tree classification learning models, random forest classification learning models and K nearest neighbor classification learning models. The decision tree classification learning model can be convenient for processing discrete and continuous data to ensure the efficiency and comprehensibility of data processing. The random forest classification learning model can have strong anti-noise ability and is convenient for processing high-dimensional data. The K nearest neighbor classification learning model can reduce the sensitivity to abnormal values ​​and does not require data input assumptions. That is, different types of classification learning models are used to more comprehensively capture abnormal behaviors in the data, and the robustness of different abnormal prediction results can be effectively improved, thereby ensuring the accuracy of the classification results.

[0028] It can be understood that the test terminal processes the initial battery data to obtain target battery data that is convenient for subsequent model processing and analysis, and can determine the corresponding coefficient factors to process the target battery data based on the data distribution diagram corresponding to each type of characteristic data, which can not only ensure the adaptability and flexibility of battery detection, but also effectively ensure that the overall distribution of the processed target battery data is normal; then, the processed target battery data can be predicted based on at least two machine learning models to obtain corresponding abnormal prediction values, so as to reduce the sensitivity to noise data and abnormal values ​​through hybrid model prediction, and can also use the diversity of the model when the battery data distribution changes to improve the overall adaptability, robustness and generalization ability, thereby ensuring the effectiveness and accuracy of battery abnormality detection.

[0029] Specifically, when performing abnormal detection of battery data, the test terminal can, but is not limited to, obtain multiple groups of initial battery data recorded by the battery test cabinet mentioned above, and the battery specifications corresponding to each group of initial battery data can be 18650 (18 mm × 65 mm) or 16430 (16 mm × 34 mm), and each group of initial battery data can be obtained by testing the same battery or different batteries of the same specifications. Here, each group of initial battery data includes at least two types of characteristic data recorded by the battery test cabinet during the battery test process, for example, it may include multiple operating temperatures of the battery in the charge and discharge state, multiple charge and discharge times, multiple charge and discharge cycle numbers, multiple charge and discharge currents, multiple charge and discharge voltages, and multiple battery capacities, and of course, it may also include multiple branch currents, multiple branch voltages, multiple battery impedances, and multiple estimated charge transfer resistances of the battery in the impedance state, and is not limited to the data types mentioned above.

[0030] Since the storage format of each set of initial battery data is generally .mat, which is a structured nested hierarchical data storage method, it is mainly used for the use and analysis of specified software (such as MATLAB software). However, this method of using specified software for data analysis has certain usage conditions, such as the need to pay for its use, and reduces the data processing efficiency to a certain extent.

[0031] Based on this, after obtaining multiple groups of initial battery data, the test terminal can also perform data conversion processing on each group of initial battery data, such as but not limited to converting each group of initial battery data into tabular data that is easy to process in Python. Compared with the existing data storage format, it can be applied to a rich third-party library (such as numpy, scipy, sklearn or matplotlib, etc.), which is sufficient to meet the vast majority of mathematical modeling needs and processing needs, and each group of processed initial battery data can be merged to obtain the target battery data. Here, the target battery data can include but is not limited to multiple operating temperatures, multiple charge and discharge times, multiple charge and discharge cycle times, multiple charge and discharge currents, multiple charge and discharge voltages, and multiple battery capacities. The corresponding tabular data, of course, can also include multiple branch currents, multiple branch voltages, multiple battery impedances, and multiple estimated charge transfer resistances. The corresponding tabular data, and is not limited to this.

[0032] As an option of the embodiment of the present application, at least two groups of initial battery data are processed to obtain target battery data, including: Based on a preset conversion function, at least two groups of initial battery data are converted respectively, and all the converted initial battery data are merged to obtain merged battery data; When missing data is detected in the merged battery data, supplementary data is obtained according to the characteristic data corresponding to the type of the missing data, and the supplementary data is filled into the merged battery data; or When missing data is detected in the merged battery data, determining preset data corresponding to the type of the missing data, and filling the preset data into the merged battery data; In the combined battery data after the filling process, statistical processing is performed on each characteristic data, and all statistical results are integrated into the combined battery data after the filling process to obtain the target battery data.

[0033] Specifically, in order to ensure the integrity and validity of the battery data, the test terminal can also obtain a preset conversion function customized by the tester. After receiving multiple sets of initial battery data (which can be read through Python), the preset conversion function is used to convert each set of initial battery data to obtain corresponding tabular data, and the merged battery data can be obtained based on all tabular data. Here, the generation method of merged battery data can be, but is not limited to, processing all tabular data according to data type. For example, the corresponding characteristic data can be classified and merged according to operating temperature, charge and discharge time, number of charge and discharge cycles, charge and discharge current, charge and discharge voltage, and battery capacity, and is not limited to this.

[0034] Next, after obtaining the target battery data, since there is a corresponding relationship between different data types, it is possible but not limited to determine whether there is missing data by checking whether the number of all feature data corresponding to each data type is consistent. It is understandable that when it is detected that the number of all feature data corresponding to any data type is lower than the number of all feature data corresponding to other data types, it indicates that there is missing data, and then supplementary data can be obtained based on all feature data corresponding to the data type of the missing data, and the supplementary data can be filled into the position corresponding to the missing data in the merged battery data.

[0035] Here, the method of obtaining supplementary data based on all feature data corresponding to the data type of the missing data can be, but is not limited to, performing mean calculation, standard deviation calculation, median query or mode query on all feature data corresponding to the data type of the missing data. Of course, the mean calculation can also be performed on at least one feature data adjacent to the missing data, but is not limited to this.

[0036] In addition, the method of obtaining supplementary data based on all characteristic data corresponding to the data type of the missing data can also be to determine the corresponding preset data according to the type of the missing data. Here, the correspondence between each data type and the corresponding preset data can be analyzed in advance by the tester, and the corresponding preset data can be filled in the position corresponding to the missing data in the merged battery data.

[0037] It should be noted that if it is detected that all missing data are of the same data type and the number of all missing data is lower than a specified threshold, it is also possible but not limited to directly eliminating all feature data corresponding to the data type of the missing data, and is not limited to this.

[0038] Next, after filling the merged battery data, the test terminal can also perform statistical processing on all characteristic data of each data type, such as but not limited to counting the number of characteristic data of each data type, the mean of characteristic data or the standard deviation of characteristic data, and can integrate the statistical results into the merged battery data after filling processing, so as to obtain the target battery data.

[0039] Step 104 : Determine corresponding coefficient factors according to the data distribution diagram corresponding to each type of characteristic data, and process the target battery data based on all coefficient factors.

[0040] Specifically, after obtaining the target battery data, the test terminal may, but is not limited to, use preset data analysis software or data analysis algorithms (both of which are well-known technical means in the art) to analyze and process all characteristic data of each type in the target battery data to obtain a corresponding data distribution graph. It is understandable that the data distribution graph may, but is not limited to, be a box plot or a histogram based on kernel density estimation, so that the test terminal can determine the corresponding coefficient factor according to the data distribution graph corresponding to each data type, and then judge whether the overall distribution of the target battery data is normal through all coefficient factors.

[0041] Here, when it is determined based on all coefficient factors that the overall distribution of the target battery data is normal, the test terminal can retain all characteristic data in the target battery data; when it is determined based on all coefficient factors that the overall distribution of the target battery data is abnormal, the test terminal can eliminate the characteristic data in the target battery data based on the quantile interval obtained by the coefficient factor.

[0042] As another option of the embodiment of the present application, according to the data distribution diagram corresponding to each type of feature data, the corresponding coefficient factor is determined, including: According to the data distribution graph corresponding to each type of feature data, the corresponding tail distribution probability is calculated; Based on the probability interval in which each tail distribution probability lies, the corresponding coefficient factor is queried in a preset probability interval-coefficient factor list; wherein the preset probability interval-coefficient factor list includes at least two groups of standard probability intervals and a standard coefficient factor corresponding to each group of standard probability intervals.

[0043] Specifically, when determining the corresponding coefficient factor according to each data distribution graph, taking the data distribution graph as a histogram based on kernel density estimation as an example, the test terminal can, but is not limited to, use one or more rectangular columns at both ends of each histogram as the tail, and obtain the tail distribution probability according to the sum of the frequencies (or frequency densities) corresponding to all rectangular columns. It can be understood that the tail distribution probability mentioned in the embodiment of the present application can be used to determine whether the corresponding data distribution graph satisfies the heavy-tailed distribution, and then the corresponding coefficient factor can be determined according to the judgment result.

[0044] Then, after obtaining the tail distribution probability corresponding to each data distribution graph, the test terminal can also query the corresponding coefficient factor in the preset probability interval-coefficient factor list according to the probability interval in which the tail distribution probability is located. For example, when the tail distribution probability is in the first probability interval, it indicates that the corresponding data distribution graph satisfies the heavy-tailed distribution. At this time, the coefficient factor queried in the preset probability interval-coefficient factor list can be, but not limited to, a specified value greater than 1.5 (for example, it can be 1.8) to more effectively capture abnormal data; when the tail distribution probability is in the second probability interval, it indicates that the corresponding data distribution graph does not satisfy the heavy-tailed distribution. At this time, the coefficient factor queried in the preset probability interval-coefficient factor list can be, but not limited to, a specified value less than 1.5 (for example, it can be 1.2) to more effectively capture abnormal data, and is not limited thereto. Here, the preset probability interval-coefficient factor list can be analyzed by the tester and can be applicable to different battery data, which includes at least two groups of standard probability intervals and standard coefficient factors corresponding to each group of standard probability intervals.

[0045] Then, after obtaining the coefficient factor, the test terminal can also determine the quantiles corresponding to all feature data of each type according to the preset quantile coefficients. For example, when the preset quantile coefficients are 10% and 90%, the feature data at the 10% position and the feature data at the 90% position of all feature data of each type can be used as quantiles, and is not limited to this.

[0046] Next, after determining the quantiles corresponding to all characteristic data of each type, the test terminal can also calculate the quantile interval corresponding to all characteristic data of each type according to the quantiles and the corresponding coefficient factors. Here, the calculation method of the quantile interval can be but is not limited to referring to the formula shown below: In the above formula, Q3 may correspond to the larger quantile in the quantile, Q1 may correspond to the smaller quantile in the quantile, K may correspond to the coefficient factor, upper may correspond to the maximum value in the quantile interval, and lower may correspond to the minimum value in the quantile interval.

[0047] Then, after determining the quantile intervals corresponding to all feature data of each type, the test terminal can also determine whether all feature data of each type are in the corresponding quantile intervals, and when it is detected that there are one or more feature data among all feature data of any type that are not in the corresponding quantile intervals, it indicates that the one or more feature data may be abnormally distributed data, and the one or more feature data can be eliminated; when it is detected that all feature data of any type are in the corresponding quantile intervals, it indicates that all feature data of this type are normally distributed data, and all feature data can be retained.

[0048] It should be noted that the traditional statistical-based abnormal data point detection method can use the interquartile range detection method to perform anomaly detection on the above-mentioned data distribution graph. The coefficient factors involved in the interquartile range detection method are generally fixed values. Compared with the method mentioned in the above embodiment, it cannot be applied to all battery data sets and cannot guarantee the accuracy of the detection results.

[0049] Also see here Figure 2 FIG. 1 is a schematic diagram showing an effect of processing target battery data based on all coefficient factors provided by an embodiment of the present application, such as Figure 2 As shown, 2a shows the box plot obtained by the traditional statistical-based abnormal data point detection method, which includes the box plots corresponding to the three types of feature data, and it can be seen that there are two types of box plots corresponding to the distribution of abnormal data (that is, the black bold lines and black bold points in 2a); 2b shows the box plot obtained based on the coefficient factor mentioned in the embodiment of the present application, which also includes the box plots corresponding to the three types of feature data, and it can be seen that there is no distribution abnormal data in the box plot corresponding to any type.

[0050] As another option of the embodiment of the present application, after processing the target battery data based on all coefficient factors, before predicting the processed target battery data based on at least two machine learning models respectively, the method further includes: According to the data heat map corresponding to all types of feature data, determine the correlation coefficient between any two types of feature data; When the correlation coefficient between any two types of feature data is in the strong correlation coefficient interval, the corresponding two types of feature data are screened; When the correlation coefficient between any two types of feature data is in a weak correlation coefficient interval, the corresponding two types of feature data are retained.

[0051] In addition to analyzing and processing any type of characteristic data, the target battery data can be further processed by analyzing the correlation between any two types of characteristic data to ensure the reliability of the target battery data and the efficiency of subsequent data processing.

[0052] Specifically, after processing the target battery data based on all coefficient factors, the test terminal can also use preset data analysis software or data analysis algorithms (both are well-known technical means in the art) to analyze and process all characteristic data of each two types in the target battery data, thereby obtaining a data heat map. Here, the data heat map can be understood as a heat map known in the art, which contains multiple positive and negative values ​​(i.e., correlation coefficients, each positive and negative value is between -1 and 1), and each positive and negative value can be used to characterize the strength of the correlation between all characteristic data of the corresponding two types. For example, when a positive and negative value is close to -1 or close to 1, it indicates that the correlation between all characteristic data of the corresponding two types is strong (for example, strong negative correlation or strong positive correlation); when a positive and negative value is close to 0, it indicates that the correlation between all characteristic data of the corresponding two types is weak (or no correlation).

[0053] It is understandable that when the correlation coefficient between any two types of feature data is detected to be in the weak correlation coefficient interval, it indicates that the correlation between all feature data of the corresponding two types is weak, and the feature data of the corresponding two types can be retained; when the correlation coefficient between any two types of feature data is detected to be in the strong correlation coefficient interval, it indicates that the correlation between all feature data of the corresponding two types is strong, and all feature data of the corresponding two types can be screened to reduce the data latitude. Here, the screening method can be, but is not limited to, screening all feature data according to a preset number of intervals. For example, the first feature data of every three feature data can be retained, and the remaining two feature data can be eliminated, and it is not limited to this.

[0054] Of course, in the embodiment of the present application, the correlation between any two types of feature data can be analyzed and judged by obtaining a scatter plot or a line plot between any two types of feature data, or the scatter plot (which can also be a line plot) and the above-mentioned data heat map can be combined for joint analysis and judgment, but no further details will be given here.

[0055] Step 106: Predict the processed target battery data based on at least two machine learning models to obtain corresponding abnormal prediction values, and obtain battery abnormality detection results based on all abnormal prediction values.

[0056] Specifically, after obtaining the processed target battery data, the test terminal may, but is not limited to, input the processed target battery data into a variety of machine learning models respectively to obtain multiple abnormal prediction values ​​through model prediction, each of which may, but is not limited to, a numerical value used to characterize abnormal battery data (for example, 0) or a numerical value used to characterize normal battery data (for example, 1), and obtain the final battery abnormality detection result by combining all abnormal prediction values.

[0057] Here, the types of machine learning models may include but are not limited to density-based spatial clustering learning model (Density-Based Spatial Clustering of Applications with Noise, DBSCAN), isolation forest learning model (Isolation Forest) and density-based anomaly detection learning model (Local Outlier Factor, LOF). Among them, the density-based spatial clustering learning model can identify each clustering cluster by combining the dense area of ​​all feature data in the target battery data, and select feature data corresponding to at least two types to cluster all clustering clusters, and determine whether there is data anomaly in the target battery data by analyzing all clustering results. You can also refer to Figure 3 A schematic diagram of the prediction effect of a machine learning model provided in an embodiment of the present application is shown, such as Figure 3 As shown, the effect diagram can be understood as a clustering effect diagram corresponding to the two data types of battery voltage and operating temperature, and the black part in the figure can correspond to the noise discrete points.

[0058] The Isolation Forest Learning Model can perform anomaly detection based on the proportion of outliers in all feature data in the target battery data. By performing anomaly detection on feature data corresponding to any two types under different proportions of outliers, it can be determined whether there are data anomalies in the target battery data. Figure 4 A schematic diagram of the prediction effect of another machine learning model provided in an embodiment of the present application is shown, such as Figure 4 As shown, the effect diagram can be understood as a diagram of the relationship between the two data types of working time and battery voltage when the proportion of outliers is 0.01. It can be seen that abnormal data points appear at the beginning of charging, and certain abnormal data points also appear at the end of charging.

[0059] The density-based anomaly detection learning model can be used to compare the local density of each feature data in the target battery data with the local density of its neighboring points to identify abnormal data. It should be noted that the density-based spatial clustering learning model, the isolation forest learning model, and the density-based anomaly detection learning model mentioned above can all be trained by testers based on multiple sets of sample data sets and the data anomaly labels corresponding to each set of sample data sets, and the model structure and related principles of each machine learning model can be well-known technical means in the field, so they will not be elaborated here.

[0060] It is understandable that after obtaining the abnormal prediction value output by each machine learning model, the test terminal can, but is not limited to, determine the battery abnormality detection result based on the mode of all abnormal prediction values. For example, when the number of abnormal prediction values ​​detected in a preset abnormal value interval among all abnormal prediction values ​​exceeds the preset number threshold, the battery abnormality detection result is determined to be abnormal battery data; when the number of abnormal prediction values ​​detected in a preset abnormal value interval among all abnormal prediction values ​​does not exceed the preset number threshold, the battery abnormality detection result is determined to be normal battery data. Here, the three machine learning models mentioned above can also be used as examples. When the number of abnormal prediction values ​​in the preset abnormal value interval exceeds one, it indicates that the battery abnormality detection result is abnormal battery data; when the number of abnormal prediction values ​​in the preset abnormal value interval does not exceed one, it indicates that the battery abnormality detection result is normal battery data, and it is not limited to this.

[0061] As another option of the embodiment of the present application, after obtaining the battery abnormality detection result according to all abnormality prediction values, the method further includes: When the battery abnormality detection result is that the battery data is abnormal, the processed target battery data is predicted based on at least two classification learning models to obtain corresponding classification prediction values; All classification prediction values ​​are concatenated to obtain a classification feature matrix, and the classification feature matrix is ​​input into the logistic regression model to obtain the abnormal classification result; Update the battery anomaly detection result to the anomaly classification result.

[0062] In order to obtain the data anomaly type more accurately, multiple classification learning models can be used to predict the target battery data separately, and all classification prediction results can be combined to obtain the final anomaly classification result.

[0063] Specifically, after determining that the battery anomaly detection result is a battery data anomaly, the test terminal can also input the processed target battery data into a variety of classification learning models to obtain multiple classification prediction values ​​through model prediction, and each classification prediction value can be, but is not limited to, a numerical value used to characterize the classification result. Here, the types of multiple classification learning models can include, but are not limited to, decision tree classification learning models, random forest classification learning models, and K-nearest neighbor classification learning models. Each classification learning model can be trained separately by the tester based on multiple groups of set sample data sets and the data classification labels corresponding to each group of sample data sets, and the model structure and related principles of each classification learning model can be technical means well known in the art, but will not be elaborated here.

[0064] Then, after obtaining the classification prediction results output by each classification learning model, the test terminal can also concatenate all classification prediction values ​​to obtain a classification feature matrix, and input the classification feature matrix into the logistic regression model to obtain the final abnormal classification result through multi-model fusion prediction. Here, the logistic regression model can be understood as a logistic regression classifier (Logistic Regression), which can also be trained by the tester based on multiple sets of classification result data sets and the classification labels corresponding to each set of classification result data sets. The model structure and related principles of the logistic regression model can be well-known technical means in the field, but will not be elaborated here.

[0065] It is understandable that the various classification learning models and logistic regression models mentioned above can be regarded as fusion multi-classification algorithms based on ensemble learning, for example, but not limited to using the three classification learning models mentioned above as primary learning models, respectively making abnormal classification predictions for battery data, and using the logistic regression model as a meta-model (or base model), and performing multi-model fusion and secondary prediction through an ensemble learning algorithm to obtain the final abnormal classification result. It should be noted that this method can ensure higher classification recognition accuracy compared to classification prediction through a separate classification learning model.

[0066] Then, after determining the final abnormal classification result, in order to facilitate the tester to intuitively understand the detection result of the battery data, the test terminal can also update the above-mentioned battery abnormality detection result to an abnormal classification result and display it to the terminal where the tester is located.

[0067] In an embodiment of the present application, by combining a decision tree classification learning model, a random forest classification learning model, and a K-nearest neighbor classification learning model to classify abnormal data, not only can the abnormal behavior in the data be captured more comprehensively, but also the overall robustness to different types of outliers can be improved; secondly, the most relevant features can be identified by using a feature selection method in each classification learning model or by using a feature selection method in a logistic regression model, thereby improving the accuracy of anomaly detection; in addition, different preset weights can be assigned to each classification learning model to further improve the overall classification performance, thereby ensuring the overall abnormal classification efficiency of the battery data.

[0068] See next Figure 5 , Figure 5 A structural schematic diagram of a battery data anomaly detection system based on a machine learning model provided in an embodiment of the present application is shown.

[0069] like Figure 5 As shown, the battery data anomaly detection system based on the machine learning model may include at least a first processing module 501, a second processing module 502 and a result prediction module 503, wherein: The first processing module 501 is used to process at least two groups of initial battery data to obtain target battery data; wherein each group of initial battery data includes at least two types of feature data; The second processing module 502 is used to determine the corresponding coefficient factors according to the data distribution diagram corresponding to each type of characteristic data, and process the target battery data based on all the coefficient factors; The result prediction module 503 is used to predict the processed target battery data based on at least two machine learning models, obtain corresponding abnormal prediction values, and obtain battery abnormality detection results according to all abnormal prediction values.

[0070] In some possible embodiments, processing at least two sets of initial battery data to obtain target battery data includes: Based on a preset conversion function, at least two groups of initial battery data are converted respectively, and all the converted initial battery data are merged to obtain merged battery data; When missing data is detected in the merged battery data, supplementary data is obtained according to the characteristic data corresponding to the type of the missing data, and the supplementary data is filled into the merged battery data; or When missing data is detected in the merged battery data, determining preset data corresponding to the type of the missing data, and filling the preset data into the merged battery data; In the combined battery data after the filling process, statistical processing is performed on each characteristic data, and all statistical results are integrated into the combined battery data after the filling process to obtain the target battery data.

[0071] In some possible embodiments, the corresponding coefficient factors are determined according to the data distribution graph corresponding to each type of characteristic data, including: According to the data distribution graph corresponding to each type of feature data, the corresponding tail distribution probability is calculated; Based on the probability interval in which each tail distribution probability lies, the corresponding coefficient factor is queried in a preset probability interval-coefficient factor list; wherein the preset probability interval-coefficient factor list includes at least two groups of standard probability intervals and a standard coefficient factor corresponding to each group of standard probability intervals.

[0072] In some possible embodiments, the target battery data is processed based on all coefficient factors, including: According to the preset quantile coefficient, the quantile corresponding to each type of characteristic data is determined, and based on the quantile corresponding to each type of characteristic data and the coefficient factor, the corresponding quantile interval is calculated; When it is detected in each type of feature data that any feature data is not within the corresponding quantile interval, the feature data is eliminated.

[0073] In some possible embodiments, after processing the target battery data based on all coefficient factors and before predicting the processed target battery data based on at least two machine learning models respectively, the method further includes: According to the data heat map corresponding to all types of feature data, determine the correlation coefficient between any two types of feature data; When the correlation coefficient between any two types of feature data is in the strong correlation coefficient interval, the corresponding two types of feature data are screened; When the correlation coefficient between any two types of feature data is in a weak correlation coefficient interval, the corresponding two types of feature data are retained.

[0074] In some possible embodiments, a battery abnormality detection result is obtained according to all abnormality prediction values, including: When the number of abnormal prediction values ​​detected in all abnormal prediction values ​​that are within a preset abnormal value interval exceeds a preset number threshold, it is determined that the battery abnormality detection result is a battery data abnormality; When the number of abnormal prediction values ​​detected in all abnormal prediction values ​​that are within a preset abnormal value interval does not exceed a preset number threshold, it is determined that the battery abnormality detection result is that the battery data is normal.

[0075] In some possible embodiments, after obtaining the battery abnormality detection result according to all abnormality prediction values, the method further includes: When the battery abnormality detection result is that the battery data is abnormal, the processed target battery data is predicted based on at least two classification learning models to obtain corresponding classification prediction values; All classification prediction values ​​are concatenated to obtain a classification feature matrix, and the classification feature matrix is ​​input into the logistic regression model to obtain the abnormal classification result; Update the battery anomaly detection result to the anomaly classification result.

[0076] Those skilled in the art can clearly understand that the technical solutions of the embodiments of the present application can be implemented with the help of software and / or hardware. The "unit" and "module" in this specification refer to software and / or hardware that can independently complete or cooperate with other components to complete specific functions, where the hardware can be, for example, a field programmable gate array (FPGA), an integrated circuit (IC), etc.

[0077] See next Figure 6 , Figure 6 A structural schematic diagram of another battery data anomaly detection system based on a machine learning model provided in an embodiment of the present application is shown.

[0078] like Figure 6 As shown, the battery data anomaly detection system 600 based on a machine learning model may include at least one processor 601 , at least one network interface 604 , a user interface 603 , a memory 605 , and at least one communication bus 602 .

[0079] The communication bus 602 may be used to realize the connection and communication among the above-mentioned components.

[0080] The user interface 603 may include buttons, and the optional user interface may also include a standard wired interface or a wireless interface.

[0081] The network interface 604 may include, but is not limited to, a Bluetooth module, an NFC module, a Wi-Fi module, etc.

[0082] Among them, the processor 601 may include one or more processing cores. The processor 601 uses various interfaces and lines to connect various parts within the battery data anomaly detection system 600 based on the machine learning model, and executes various functions and processes data of the battery data anomaly detection system 600 based on the machine learning model by running or executing instructions, programs, code sets or instruction sets stored in the memory 605, and calling data stored in the memory 605. Optionally, the processor 601 can be implemented in at least one hardware form of DSP, FPGA, and PLA. The processor 601 can integrate one or a combination of CPU, GPU, modem, etc. Among them, the CPU mainly processes the operating system, user interface, and application program; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 601, and it can be implemented by a single chip.

[0083] The memory 605 may include a RAM or a ROM. Optionally, the memory 605 includes a non-transitory computer-readable medium. The memory 605 may be used to store instructions, programs, codes, code sets or instruction sets. The memory 605 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 605 may optionally be at least one storage device located away from the aforementioned processor 601. As Figure 6 As shown, the memory 605 as a computer storage medium may include an operating system, a network communication module, a user interface module, and a battery data anomaly detection application based on a machine learning model.

[0084] Specifically, the processor 601 may be used to call a battery data anomaly detection application based on a machine learning model stored in the memory 605, and specifically perform the following operations: Processing at least two groups of initial battery data to obtain target battery data; wherein each group of initial battery data includes at least two types of characteristic data; According to the data distribution diagram corresponding to each type of characteristic data, the corresponding coefficient factors are determined, and the target battery data is processed based on all the coefficient factors; Based on at least two machine learning models, the processed target battery data is predicted respectively to obtain corresponding abnormal prediction values, and the battery abnormality detection result is obtained according to all the abnormal prediction values.

[0085] In some possible embodiments, processing at least two sets of initial battery data to obtain target battery data includes: Based on a preset conversion function, at least two groups of initial battery data are converted respectively, and all the converted initial battery data are merged to obtain merged battery data; When missing data is detected in the merged battery data, supplementary data is obtained according to the characteristic data corresponding to the type of the missing data, and the supplementary data is filled into the merged battery data; or When missing data is detected in the merged battery data, determining preset data corresponding to the type of the missing data, and filling the preset data into the merged battery data; In the combined battery data after the filling process, statistical processing is performed on each characteristic data, and all statistical results are integrated into the combined battery data after the filling process to obtain the target battery data.

[0086] In some possible embodiments, the corresponding coefficient factors are determined according to the data distribution graph corresponding to each type of characteristic data, including: According to the data distribution graph corresponding to each type of feature data, the corresponding tail distribution probability is calculated; Based on the probability interval in which each tail distribution probability lies, the corresponding coefficient factor is queried in a preset probability interval-coefficient factor list; wherein the preset probability interval-coefficient factor list includes at least two groups of standard probability intervals and a standard coefficient factor corresponding to each group of standard probability intervals.

[0087] In some possible embodiments, the target battery data is processed based on all coefficient factors, including: According to the preset quantile coefficient, the quantile corresponding to each type of characteristic data is determined, and based on the quantile corresponding to each type of characteristic data and the coefficient factor, the corresponding quantile interval is calculated; When it is detected in each type of feature data that any feature data is not within the corresponding quantile interval, the feature data is eliminated.

[0088] In some possible embodiments, after processing the target battery data based on all coefficient factors and before predicting the processed target battery data based on at least two machine learning models respectively, the method further includes: According to the data heat map corresponding to all types of feature data, determine the correlation coefficient between any two types of feature data; When the correlation coefficient between any two types of feature data is in the strong correlation coefficient interval, the corresponding two types of feature data are screened; When the correlation coefficient between any two types of feature data is in a weak correlation coefficient interval, the corresponding two types of feature data are retained.

[0089] In some possible embodiments, a battery abnormality detection result is obtained according to all abnormality prediction values, including: When the number of abnormal prediction values ​​detected in all abnormal prediction values ​​that are within a preset abnormal value interval exceeds a preset number threshold, it is determined that the battery abnormality detection result is a battery data abnormality; When the number of abnormal prediction values ​​detected in all abnormal prediction values ​​that are within a preset abnormal value interval does not exceed a preset number threshold, it is determined that the battery abnormality detection result is that the battery data is normal.

[0090] In some possible embodiments, after obtaining the battery abnormality detection result according to all abnormality prediction values, the method further includes: When the battery abnormality detection result is that the battery data is abnormal, the processed target battery data is predicted based on at least two classification learning models to obtain corresponding classification prediction values; All classification prediction values ​​are concatenated to obtain a classification feature matrix, and the classification feature matrix is ​​input into the logistic regression model to obtain the abnormal classification result; Update the battery anomaly detection result to the anomaly classification result.

[0091] The present application also provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of the above method are implemented. The computer-readable storage medium may include, but is not limited to, any type of disk, including a floppy disk, an optical disk, a DVD, a CD-ROM, a micro drive, and a magneto-optical disk, a ROM, a RAM, an EPROM, an EEPROM, a DRAM, a VRAM, a flash memory device, a magnetic card or an optical card, a nanosystem (including a molecular memory IC), or any type of medium or device suitable for storing instructions and / or data.

[0092] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0093] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0094] In the several embodiments provided in the present application, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interfaces, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0095] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0096] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

Claims

1. A battery data anomaly detection method based on a machine learning model, characterized in that: include: Processing at least two groups of initial battery data to obtain target battery data; wherein each group of the initial battery data includes at least two types of characteristic data; Determine corresponding coefficient factors according to the data distribution diagram corresponding to each type of the characteristic data, and process the target battery data based on all the coefficient factors; The processed target battery data is predicted based on at least two machine learning models to obtain corresponding abnormal prediction values, and a battery abnormality detection result is obtained based on all the abnormal prediction values.

2. The method according to claim 1, characterized in that The processing of at least two sets of initial battery data to obtain target battery data includes: Performing conversion processing on at least two groups of initial battery data respectively based on a preset conversion function, and merging all the initial battery data after the conversion processing to obtain merged battery data; When missing data is detected in the merged battery data, supplementary data is obtained according to the characteristic data corresponding to the type of the missing data, and the supplementary data is filled into the merged battery data; or When missing data is detected in the merged battery data, determining preset data corresponding to the type of the missing data, and filling the merged battery data with the preset data; In the combined battery data after the filling process, statistical processing is performed on each of the characteristic data, and all statistical results are integrated into the combined battery data after the filling process to obtain target battery data.

3. The method according to claim 1, characterized in that Determining corresponding coefficient factors according to the data distribution diagram corresponding to each type of the characteristic data includes: Calculate the corresponding tail distribution probability according to the data distribution graph corresponding to each type of the characteristic data; Based on the probability interval in which each of the tail distribution probabilities is located, the corresponding coefficient factor is queried in the preset probability interval-coefficient factor list; wherein the preset probability interval-coefficient factor list includes at least two groups of standard probability intervals, and the standard coefficient factor corresponding to each group of the standard probability intervals.

4. The method according to claim 1 or 3, characterized in that: The processing of the target battery data based on all the coefficient factors includes: According to a preset quantile coefficient, the quantile corresponding to each type of the characteristic data is determined, and based on the quantile corresponding to each type of the characteristic data and the coefficient factor, the corresponding quantile interval is calculated; When it is detected that any one of the feature data of each type is not within the corresponding quantile interval, the feature data is eliminated.

5. The method according to claim 1, characterized in that After processing the target battery data based on all the coefficient factors and before predicting the processed target battery data based on at least two machine learning models respectively, the method further includes: Determine the correlation coefficient between any two types of the feature data according to the data heat map corresponding to all types of the feature data; When the correlation coefficient between any two types of the feature data is in a strong correlation coefficient interval, screening the corresponding two types of the feature data; When the correlation coefficient between any two types of the feature data is in a weak correlation coefficient interval, the corresponding two types of the feature data are retained.

6. The method according to claim 1, characterized in that The battery abnormality detection result is obtained according to all the abnormality prediction values, including: When the number of the abnormal prediction values ​​detected in all the abnormal prediction values ​​that are within the preset abnormal value interval exceeds a preset number threshold, determining that the battery abnormality detection result is a battery data abnormality; When the number of abnormal prediction values ​​detected in all the abnormal prediction values ​​that are within the preset abnormal value interval does not exceed the preset number threshold, it is determined that the battery abnormality detection result is that the battery data is normal.

7. The method according to claim 1 or 6, characterized in that: After obtaining the battery abnormality detection result according to all the abnormality prediction values, the method further includes: When the battery abnormality detection result is battery data abnormality, predicting the processed target battery data based on at least two classification learning models respectively to obtain corresponding classification prediction values; All the classification prediction values ​​are concatenated to obtain a classification feature matrix, and the classification feature matrix is ​​input into a logistic regression model to obtain an abnormal classification result; The battery abnormality detection result is updated to the abnormality classification result.

8. A battery data anomaly detection system based on a machine learning model, characterized in that: include: A first processing module, configured to process at least two groups of initial battery data to obtain target battery data; wherein each group of the initial battery data includes at least two types of feature data; A second processing module, configured to determine corresponding coefficient factors according to a data distribution diagram corresponding to each type of the characteristic data, and process the target battery data based on all the coefficient factors; The result prediction module is used to predict the processed target battery data based on at least two machine learning models, obtain corresponding abnormal prediction values, and obtain battery abnormality detection results based on all the abnormal prediction values.

9. A battery data anomaly detection system based on a machine learning model, characterized in that: including a processor and a memory; The processor is connected to the memory; The memory is used to store executable program code; The processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to execute the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed on a computer or a processor, the computer or the processor executes the steps of the method according to any one of claims 1 to 7.

Images