Data construction methods, apparatus, battery pack health status detection methods and equipment

Abstract

This application relates to a data construction method, apparatus, battery pack health status detection method, and device. The method includes: dividing the data of each normal battery pack into multiple charge / discharge segments; extracting features from the charge / discharge segments of the multiple normal battery packs to obtain feature vectors for the multiple charge / discharge segments of each normal battery pack; determining the probability density distribution of all feature vectors of the normal battery pack based on a kernel function; determining the inverse probability density distribution of abnormal battery packs based on the probability density distribution; and sampling the inverse probability density distribution to obtain abnormal battery pack data. Using this application can effectively address the problem of scarce abnormal samples, expand the number of abnormal samples, and improve the imbalance of training feature distribution.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a data construction method, apparatus, battery pack health status detection method and device. Background Technology

[0002] With the rapid development of the new energy industry, power storage systems have shown broad application prospects and huge market potential. Among the many energy storage technologies, lithium iron phosphate (LFP) batteries have become the mainstream choice in the field of power storage due to their high energy density, long cycle life, low self-discharge rate, and good safety and environmental friendliness.

[0003] Batteries (including lithium iron phosphate batteries) experience gradual performance degradation during actual operation due to factors such as electrochemical aging, environmental stress, and operating conditions. This degradation not only affects the system's energy output and operating efficiency but can also lead to safety risks such as thermal runaway in severe cases. To scientifically assess the degree of battery performance degradation, the industry widely adopts State of Health (SOH) as a key characterization indicator.

[0004] Currently, the recognition accuracy of training models relies on large datasets, including both large sets of positive and negative samples. Typically, compared to normal battery pack data (positive samples), abnormal negative sample data is much harder to obtain. However, the difference in quantity between the positive and negative sample sets leads to data imbalance between different categories, which affects the recognition accuracy of the training model. Therefore, how to construct abnormal negative sample data has become a pressing technical problem to be solved. Summary of the Invention

[0005] Therefore, it is necessary to provide a data construction method, apparatus, battery pack health status detection method and device to address the above-mentioned technical problems. This method and device can fully utilize the distribution information of normal pack data to construct abnormal pack data, effectively address the problem of scarce abnormal samples, expand the number of abnormal samples, and improve the imbalance of training feature distribution.

[0006] Firstly, this application provides a data construction method, which includes:

[0007] The data from each normal packet is divided into multiple charge / discharge segments;

[0008] Feature extraction is performed on the charging and discharging segments of multiple normal packets to obtain feature vectors of multiple charging and discharging segments of each normal packet.

[0009] The probability density distribution of all feature vectors of the normal package is determined based on the kernel function;

[0010] Determine the inverse probability density distribution of the abnormal packet based on the probability density distribution;

[0011] Sampling is performed on the inverse probability density distribution to obtain abnormal packet data.

[0012] In one embodiment, the method further includes:

[0013] Based on the battery voltage difference of multiple charging segments in each normal pack data, determine k typical charging segments for each normal pack data.

[0014] Based on the battery voltage difference of multiple discharge segments in each normal pack data, determine k typical discharge segments for each normal pack data.

[0015] k typical charging segments and k typical discharging segments are used as the charging and discharging segments for feature extraction, and each charging and discharging segment is either a charging segment or a discharging segment.

[0016] In one embodiment, feature extraction is performed on the charge-discharge segments of multiple normal packets to obtain feature vectors for multiple charge-discharge segments of each normal packet, including:

[0017] The maximum voltage of all cells in the battery pack at each sampling time of each charge-discharge segment is obtained. Based on the maximum voltage of all sampling times of each charge-discharge segment, the maximum single cell voltage of each charge-discharge segment is determined. The first feature is the maximum single cell voltage of the segment.

[0018] The minimum voltage of all cells in the battery pack at each sampling time of each charge-discharge segment is obtained. Based on the minimum voltage at all sampling times of each charge-discharge segment, the minimum single cell voltage of each charge-discharge segment is determined. The second feature is the minimum single cell voltage of the segment.

[0019] The first mean or the first variance of the current value of the battery pack at all sampling times of each charge-discharge segment is obtained, and the third feature is the first mean or the first variance.

[0020] The second mean or second variance of the state of charge of the battery pack at all sampling times of each charge-discharge segment is obtained, and the fourth feature is the second mean or second variance.

[0021] The third mean or third difference of the temperature of the battery pack at all sampling times of each charge and discharge segment is obtained, and the fifth feature is the third mean or third difference.

[0022] The feature vector of each charge / discharge segment is composed of at least two of the first to fifth features.

[0023] In one embodiment, the probability density distribution of all feature vectors of the normal package is determined based on a kernel function. The expression for calculating the probability density distribution is as follows:

[0024]

[0025] Wherein, the feature vector z j Let be the independent variable of the probability density distribution. For the eigenvector z j The probability density value of z; i Let represent the feature vector of the i-th charging segment or the i-th discharging segment of any normal packet data, n be the total number of charging or discharging segments of all normal packets, K be the kernel function, d be the dimension of the feature vector, and h be the window parameter of the kernel function K.

[0026] In one embodiment, the inverse probability density distribution belonging to the anomalous packet is determined based on the probability density distribution. The expression for calculating the inverse probability density distribution is as follows:

[0027]

[0028] in, For the eigenvector z j The probability density distribution, where n is the total number of charging or discharging segments of all normal packets. This represents the inverse probability density distribution of the abnormal packets.

[0029] In one embodiment, sampling the inverse probability density distribution to obtain abnormal packet data includes:

[0030] Calculate the cumulative distribution function based on the inverse probability density distribution;

[0031] Find the inverse function of the cumulative distribution function;

[0032] Samples are taken from a standard uniform distribution to obtain multiple horizontal axis values ​​for the standard uniform distribution;

[0033] Substitute each horizontal axis value into the inverse function to obtain the anomaly packet data;

[0034] Wherein, cumulative distribution function The calculation expression is:

[0035]

[0036] in, Let z be the inverse probability density distribution of the anomalous packets, and u be the feature vector z. j (z) j1 ,z j2 ,z j3 ,z j4 ,z j5 ) T Any value in the range.

[0037] In one embodiment, before dividing the data of each normal packet into multiple charge / discharge segments, the data construction method further includes data cleaning of any type of packet, which can be a normal packet or an abnormal packet. The data cleaning process specifically includes:

[0038] If the sample data in sample package A corresponding to each type of package meets the outlier condition, then sample package A of that type of package is cleared; if a sample package belonging to a type of package has missing sample data at time t, then the sample data of that sample package at time t is filled in using the sample data of the time adjacent to time t.

[0039] Among them, sample package A is one of the following: battery pack current sample data, battery pack state of charge sample data, and cell voltage sample data corresponding to any type of package; the outlier condition is: the absolute difference between the sample data of two adjacent time points of sample package A is greater than a preset threshold.

[0040] In one embodiment, the charge / discharge segment is either a charging segment or a discharging segment, and multiple charge / discharge segments are divided for each normal packet data, including:

[0041] According to a preset time interval, each normal packet data is divided into segments to obtain multiple segment data;

[0042] If the amount of discharge in a data segment is greater than the amount of charge during the time from the start time to the end time of the segment, then the data segment is determined to be a discharge segment.

[0043] If the discharge amount of a data segment is less than the charging amount during the time from the start time of the segment to the end time of the segment, then the data segment is determined to be a charging segment.

[0044] Secondly, this application also provides a method for detecting the health status of a battery pack, the method comprising:

[0045] Obtain multiple normal packet data and construct a method to obtain multiple abnormal packet data based on the data from any item in the first aspect;

[0046] Training features are selected based on multiple normal and multiple abnormal packet data to determine the training features input into the training model.

[0047] A state recognition model is obtained by training the model based on the training features.

[0048] The state recognition model is used to detect the state of the battery pack under test and determine whether the battery pack is normal or abnormal.

[0049] Thirdly, this application also provides a data construction apparatus, which includes:

[0050] The segmentation module is used to divide each normal packet data into multiple charge / discharge segments;

[0051] The feature extraction module is used to extract features from the charge and discharge segments of multiple normal packets to obtain feature vectors of multiple charge and discharge segments of each normal packet.

[0052] The probability density determination module is used to determine the probability density distribution of all feature vectors of the normal package based on the kernel function.

[0053] The distribution determination module is used to determine the inverse probability density distribution belonging to the abnormal packet based on the probability density distribution;

[0054] The data acquisition module is used to sample the inverse probability density distribution to obtain abnormal packet data.

[0055] Fourthly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of either the first or second aspect.

[0056] Fifthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of either the first or second aspect.

[0057] In a sixth aspect, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of either the first or second aspect.

[0058] The aforementioned data construction method, apparatus, battery pack health status detection method, and device divide the data of each normal pack into multiple charge-discharge segments; extract features from the charge-discharge segments of multiple normal packs to obtain feature vectors for multiple charge-discharge segments of each normal pack; determine the probability density distribution of all feature vectors of the normal pack based on a kernel function; determine the inverse probability density distribution of the abnormal pack based on the probability density distribution; and sample the inverse probability density distribution to obtain the abnormal pack data. This embodiment of the application fully utilizes the distribution information of the normal pack data to construct the abnormal pack data, which can effectively address the problem of scarce abnormal samples, expand the number of abnormal samples, and improve the imbalance of training feature distribution. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 This is a flowchart illustrating the training feature selection method in one embodiment;

[0061] Figure 2 This is a flowchart illustrating the process of dividing a process into multiple charging and discharging segments in one embodiment.

[0062] Figure 3 This is a flowchart illustrating the steps for determining typical charging and discharging segments in one embodiment.

[0063] Figure 4 This is a flowchart illustrating the steps for screening statistical features in one embodiment;

[0064] Figure 5a This is one of the schematic diagrams illustrating the characteristic distribution of normal and abnormal packets in one embodiment;

[0065] Figure 5b This is a second schematic diagram illustrating the characteristic distribution of normal and abnormal packets in one embodiment;

[0066] Figure 6 This is a flowchart illustrating a data construction method in one embodiment;

[0067] Figure 7 This is a flowchart illustrating a battery pack health status detection method in one embodiment;

[0068] Figure 8 This is a structural block diagram of a data construction apparatus in one embodiment;

[0069] Figure 9 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0070] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0071] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0072] With the rapid development of the new energy industry, lithium iron phosphate (LFP) batteries have become the mainstream choice in the field of power storage. During actual operation, batteries (including LFP batteries) experience gradual performance degradation due to factors such as electrochemical aging, environmental stress, and operating conditions. This degradation not only affects the system's energy output and operating efficiency but can also, in severe cases, lead to safety risks such as thermal runaway.

[0073] Currently, cloud platforms have accumulated massive amounts of battery pack data under real-world operating conditions. This data can be used to train a health status detection model, enabling timely identification of battery packs with poor health. Therefore, how to construct training data has become a pressing technical problem to be solved.

[0074] To address the aforementioned problems, this application provides a data construction method that divides each normal packet data into multiple charge / discharge segments; extracts features from the charge / discharge segments of the multiple normal packets to obtain feature vectors for each normal packet; determines the probability density distribution of all feature vectors of the normal packets based on a kernel function; determines the inverse probability density distribution of packets belonging to abnormal packets based on the probability density distribution; and samples the inverse probability density distribution to obtain abnormal packet data. This application fully utilizes the distribution information of normal packet data to construct abnormal packet data, effectively addressing the problem of scarce abnormal samples, expanding the number of abnormal samples, and improving the imbalance of training feature distribution.

[0075] In one exemplary embodiment, such as Figure 1 As shown, a training feature selection method is provided. This method is illustrated using a server as an example. It is understood that this method can also be applied to a terminal, or to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. This training feature selection method may include the following steps 101 to 105.

[0076] Step 101: Obtain data for multiple battery packs.

[0077] The data from multiple battery packs includes data from multiple normal packs representing batteries that are functioning normally, and data from multiple abnormal packs representing batteries that are malfunctioning. For a single battery pack, the data includes individual cell voltage, pack current, SOC (State of Charge), temperature, and charge level at each sampling time.

[0078] During operation, an electrical device comprising N battery packs uploads battery pack data to a server for storage. The N battery packs can include normal packs and abnormal packs. Battery packs with normal battery performance are designated as normal packs, while those with abnormal battery performance are designated as abnormal packs. Accordingly, data for each battery pack is stored on the server as either normal pack data or abnormal pack data. Normal battery performance is defined as follows: within a preset temperature range, the change in individual cell voltage and the change in battery capacity exhibit the same trend. For example, within the preset temperature range, the difference in the maximum voltage of individual cells in the battery pack at two adjacent sampling times is proportional to the difference in battery capacity at those two sampling times. Conversely, abnormal battery performance is defined as follows: within the preset temperature range, the change in individual cell voltage and the change in battery capacity do not exhibit the same trend.

[0079] Step 102: Divide the data of each battery pack into multiple charge and discharge segments.

[0080] The charging and discharging segment includes a charging segment and a discharging segment; for a charging segment, the charging capacity of the charging segment is greater than the discharging capacity; for a discharging segment, the discharging capacity of the discharging segment is greater than the charging capacity.

[0081] Step 103: Extract features from each charge / discharge segment to obtain multiple statistical features of the associated battery parameter characteristics of each battery pack.

[0082] Battery parameter characteristics may include, but are not limited to, cell voltage, battery pack charge variation, and battery pack temperature. Statistical characteristics may include, but are not limited to, the maximum and minimum cell voltages within a segment, charge variation, skewness of the maximum and minimum cell voltages within a segment, and the ratio of relative voltage difference to relative charge difference. Kurtosis is a statistic that measures the sharpness of a data distribution. Skewness is a statistic used to describe the asymmetry of a data distribution. It should be noted that statistical characteristics are not limited to the examples above, and other characteristics may be included in practical applications.

[0083] For each charge / discharge segment of data for each battery pack, extract the aforementioned statistical features for that charge / discharge segment.

[0084] Step 104: For each statistical feature associated with each parameter characteristic, perform feature distribution statistics on all normal packets to obtain the first feature distribution, and perform feature distribution statistics on all abnormal packets to obtain the second feature distribution.

[0085] For example, the battery parameter characteristic is cell voltage, and the statistical feature associated with cell voltage is the maximum single-cell voltage of a segment. For all normal pack data, multiple feature values ​​of the maximum single-cell voltage of multiple segments in all charging and discharging segments of all normal pack data are obtained, and a first feature distribution is obtained by performing distribution statistics based on the multiple feature values. For all abnormal pack data, multiple feature values ​​of the maximum single-cell voltage of multiple segments in all charging and discharging segments of all abnormal pack data are obtained, and a second feature distribution is obtained by performing distribution statistics based on the multiple feature values. For other statistical features, the corresponding first feature distribution and corresponding second feature distribution are obtained in the same way.

[0086] Step 105: Based on the first feature distribution and the second feature distribution that are associated with the same parameter characteristics, determine the statistical features that show significant differences between normal packets and abnormal packets as training features for training the model.

[0087] For each statistical feature, calculate the feature difference value between the first feature distribution and the second feature distribution. If the feature difference value is greater than or equal to a preset difference value, the difference between normal and abnormal packets is determined to be significant, and this statistical feature can be used as a training feature for the training model. If the feature difference value is less than the preset difference value, the difference between normal and abnormal packets is determined to be insignificant, and this statistical feature is not used as a training feature for the training model.

[0088] For example, for a certain statistical feature f, the feature difference value between the first feature distribution and the second feature distribution of the statistical feature f is calculated. If the feature difference value is greater than a preset difference value, the statistical feature f is used as a training feature; if the feature difference value is not greater than the preset difference value, the statistical feature f is not used as a training feature.

[0089] In the above embodiments, feature selection based on the feature distribution differences between normal battery packs with normal battery performance and abnormal battery packs with abnormal battery performance can improve the discriminative power of training features, thereby improving the model training effect and enabling the trained model to efficiently and accurately detect the health status of battery packs.

[0090] In order to remove outlier data points in each battery pack, the battery pack data needs to be cleaned before extracting statistical features.

[0091] In an exemplary embodiment, the training feature selection method of this application further includes data cleaning of any type of package, where any type of package can be a normal package or an abnormal package. The data cleaning process specifically includes:

[0092] If the sample data in sample package A corresponding to each type of package meets the outlier condition, then sample package A of that type of package is cleared; if a sample package belonging to a certain type of package has missing sample data at time t, then the sample data of that sample package at time t is supplemented using the sample data of the time adjacent to time t.

[0093] Among them, sample package A is one of the following: battery pack current sample data, battery pack state of charge sample data, and cell voltage sample data corresponding to any type of package; the outlier condition is: the absolute difference between the sample data of two adjacent time points of sample package A is greater than a preset threshold.

[0094] It's worth noting that for a battery pack, sample data such as battery pack current, battery pack SOC, and the voltage of each individual cell can be collected at any given moment. Therefore, a battery pack corresponds to at least multiple sample packets consisting of battery pack current data from multiple sampling moments, multiple sample packets consisting of battery pack SOC data from multiple sampling moments, and multiple sample packets consisting of individual cell voltage data from multiple sampling moments. Each sample packet records the sampling moment and the sample data value for each data point. For example, for sample packet A storing battery pack current, sample packet A contains multiple battery pack current sample data points, and any sample data point P... t Both have a sampling time t and the battery pack current value sampled at sampling time t.

[0095] In one embodiment, for a battery pack B, in a sample pack A composed of battery pack current data at multiple sampling times, if the absolute difference between the battery pack current value at sampling time t and the battery pack current value at sampling time t+1 is greater than a preset threshold, then sample pack A is cleared.

[0096] In one embodiment, for a battery pack B, in a sample pack A composed of battery pack current data from multiple sampling times, if the battery pack current value at sampling time t and the battery pack current value at sampling time t+2 both exist, but the sample pack A lacks the battery pack current value at sampling time t+1, then the battery pack current value I2 at sampling time t+1 is supplemented using the battery pack current value I1 at sampling time t and the battery pack current value I3 at sampling time t+2. That is, the battery pack current value I2 is the average of the battery pack current values ​​I1 and I3.

[0097] In the above embodiments, data cleaning processing for each type of package data can improve the quality of battery pack data. Furthermore, cleaning normal package data can construct high-quality abnormal package data based on normal package data, thereby improving the quality of training features and enhancing the training efficiency and detection performance of the model.

[0098] In an exemplary embodiment, a charge / discharge segment is either a charging segment or a discharging segment, referring to... Figure 2 For step 102: dividing the data of each battery pack into multiple charge and discharge segments, it may include the following steps 201 to 202.

[0099] Step 201: Divide the data of each battery pack into segments according to a preset time interval to obtain multiple segments of data.

[0100] For example, based on a preset time interval Δt, the start time of the first segment is determined as t1, and the end time as t2; the start time of the second segment is t2, and the end time as t3; and so on, the start and end times of multiple segments can be determined. The battery pack data between t1 and t2 is divided into the first segment data; the battery pack data between t2 and t3 is divided into the second segment data. And so on, multiple segment data can be obtained.

[0101] Step 202: If the discharge capacity of a data segment is greater than the charging capacity during the time from the start time to the end time of the segment, then the data segment is determined to be a discharge segment; if the discharge capacity of a data segment is less than the charging capacity during the time from the start time to the end time of the segment, then the data segment is determined to be a charging segment.

[0102] For a given data segment, both charging and discharging may occur simultaneously. Therefore, the amount of electricity discharged from the start to the end of the segment is compared with the amount of electricity charged during the same period. If the amount of electricity discharged is greater than the amount of electricity charged, it indicates an overall trend of discharging, and the data segment is determined to be a discharging segment; if the amount of electricity discharged is less than the amount of electricity charged, it indicates an overall trend of charging, and the data segment is determined to be a charging segment.

[0103] It's worth noting that a data segment includes the state of charge (SOC) at multiple sampling times. If the SOC at a later sampling time is greater than the SOC at a previous sampling time, then the battery was charged between the two sampling times. The absolute difference in SOC between these two sampling times is determined as the stage charge amount between the two adjacent sampling times. Conversely, if the SOC at a later sampling time is less than the SOC at a previous sampling time, then the battery was discharged between the two sampling times. The absolute difference in SOC between these two sampling times is determined as the stage discharge amount between the two adjacent sampling times. For a single segment, the sum of all stage charge amounts (SUM1) and the sum of all stage discharge amounts (SUM2) at all adjacent sampling times are calculated. If the sum of stage charge amounts (SUM1) is greater than the sum of stage discharge amounts (SUM2), then the segment is a charging segment; if the sum of stage charge amounts (SUM1) is less than the sum of stage discharge amounts (SUM2), then the segment is a discharging segment.

[0104] In this embodiment, after dividing each normal or abnormal battery pack into charging and discharging segments, the aforementioned statistical features are extracted for each charging or discharging segment. For step 103: extracting features from each charging and discharging segment to obtain multiple statistical features of each battery pack's associated multiple battery parameter characteristics, this may include:

[0105] 1) Obtain the maximum voltage of all cells in the battery pack at each sampling time of each charge-discharge segment, and determine the maximum single cell voltage of each charge-discharge segment based on the maximum voltage at all sampling times of each charge-discharge segment.

[0106] The formula for calculating the maximum single-cell voltage of a segment is as follows:

[0107]

[0108] in, This represents the maximum single-cell voltage of the segment. denoted as the maximum voltage of all cells in the battery pack at the i-th sampling time in the charge / discharge segment, and m is the number of sampling times in the charge / discharge segment.

[0109] 2) Obtain the minimum voltage of all cells in the battery pack at each sampling time of each charge / discharge segment, and determine the minimum single cell voltage of each charge / discharge segment based on the minimum voltage at all sampling times of each charge / discharge segment.

[0110] The expression for calculating the minimum unit voltage of a segment is as follows:

[0111]

[0112] in, The minimum unit voltage of the segment. Let m be the minimum voltage of all cells in the battery pack at the i-th sampling time in the charge / discharge segment, and m be the number of sampling times in the charge / discharge segment.

[0113] 3) Determine the change in charge for each charge / discharge segment based on the charge and discharge amounts at the start and end of each segment.

[0114] The expression for calculating the change in electricity is as follows:

[0115]

[0116] Where dE is the change in electrical charge, and abs() is the absolute value. The amount of charge at the moment the segment ends. The amount of charge at the start of the segment. This represents the discharge charge at the end of the segment. This represents the discharge charge at the start of the segment.

[0117] 4) Calculate the voltage difference between the maximum single-cell voltage characteristic and the minimum single-cell voltage characteristic of each charge / discharge segment, and calculate the ratio between the voltage difference and the change in charge to obtain the ratio of the relative voltage difference to the relative charge difference for each charge / discharge segment.

[0118]

[0119] Where dVE is the ratio of the relative voltage difference to the relative charge difference in the charge / discharge segment, and dE is the change in charge in the charge / discharge segment. This represents the maximum single-cell voltage of the segment. This represents the minimum unit voltage of the segment.

[0120] 5) Based on the sampled values ​​at each sampling time of each charge / discharge segment and the mean of the sampled values ​​at all sampling times of each charge / discharge segment, obtain the kurtosis and skewness values ​​for each charge / discharge segment. The sampled value at each sampling time of a charge / discharge segment is either the minimum voltage of all cells in the battery pack to which that segment belongs at that sampling time, or the maximum voltage of all cells in the battery pack to which that segment belongs at that sampling time. Kurtosis is a statistic that measures the angularity of the data distribution. Skewness is a statistic used to describe the asymmetry of the data distribution; the sign of the skewness value indicates the direction of skewness, with positive values ​​indicating right skewness and negative values ​​indicating left skewness.

[0121] The formula for calculating kurtosis is as follows:

[0122]

[0123] This involves calculating the kurtosis value for each discharge segment and each charging segment. Taking the calculation of the kurtosis value for any discharge segment as an example... Let x be the mean of the sampled values ​​at all sampling times in the discharge segment for which the kurtosis value is to be solved. i Let m be the sampled value at the i-th sampling moment within the discharge segment for which the kurtosis value is to be solved, and m be the total number of sampling moments in the discharge segment for which the kurtosis value is to be solved. Understandably, the sampled value can be associated with the minimum or maximum voltage of all cells. Therefore, for any discharge segment, the kurtosis value associated with the minimum voltage of all cells and the kurtosis value associated with the maximum voltage of all cells can be calculated.

[0124] The expression for calculating the skewness value is as follows:

[0125]

[0126] This involves calculating the skewness value for each discharge segment and each charging segment. Taking the calculation of the kurtosis value for any discharge segment as an example... Let x be the mean of the sampled values ​​at all sampling times in the discharge segment for which the skewness value is to be solved. i Let m be the sampled value at the i-th sampling moment within the discharge segment for which the skewness value is to be solved, and m be the total number of sampling moments in the discharge segment for which the skewness value is to be solved. Understandably, the sampled value can be associated with the minimum or maximum voltage of all cells. Therefore, for any discharge segment, the skewness value associated with the minimum voltage of all cells and the skewness value associated with the maximum voltage of all cells can be calculated.

[0127] In some embodiments, feature extraction for each charge / discharge segment includes: feature extraction for k typical charging segments and k typical discharging segments.

[0128] In the above embodiments, feature extraction is performed on each charge-discharge segment to obtain multiple statistical features, which provides data support for subsequent determination of feature distribution and selection of training features.

[0129] To better filter out the distinguishing features between abnormal and normal packets, the training feature filtering method in this application can perform typical charge-discharge segment filtering before obtaining multiple statistical features, such as... Figure 3 As shown, the specific steps include the following:

[0130] Step 301: Determine k typical charging segments for each battery pack based on the battery voltage differences of multiple charging segments in each battery pack data.

[0131] For each charging segment, the difference between the battery voltage at the end of the segment and the battery voltage at the beginning of the segment is calculated to obtain the battery voltage difference for that charging segment. The battery voltage differences of multiple charging segments are arranged in descending order, and the top k segments are selected as typical charging segments.

[0132] Step 302: Determine k typical discharge segments for each battery pack based on the battery voltage differences of multiple discharge segments in each battery pack data.

[0133] For each discharge segment, the difference between the battery voltage at the start of the segment and the battery voltage at the end of the segment is calculated to obtain the battery voltage difference for that discharge segment. The battery voltage differences of multiple discharge segments are arranged in descending order, and the top k segments are selected as typical discharge segments.

[0134] Step 303: Use k typical charging segments and k typical discharging segments as multiple charging and discharging segments for extracting statistical features. Each charging and discharging segment is either a charging segment or a discharging segment.

[0135] In practical applications, statistical features can be extracted from k typical charging segments and k typical discharging segments in battery pack data to determine the distribution of statistical features.

[0136] Understandably, segments with obvious charging or discharging trends can be selected based on the battery voltage difference. Statistical features can be extracted from these segments with obvious trends, which are more conducive to training the model.

[0137] In some embodiments, k typical charging segments of each normal package data are determined based on the battery voltage differences of multiple charging segments of each normal package data; k typical discharging segments of each normal package data are determined based on the battery voltage differences of multiple discharging segments of each normal package data; the k typical charging segments and k typical discharging segments are used as charge-discharge segments for feature extraction, where each charge-discharge segment is either a charging segment or a discharging segment, i.e., the statistical features extracted in step 103 above are performed on the k typical charging segments and k typical discharging segments. Similarly, k typical charging segments and k typical discharging segments can be determined for abnormal package data in the same manner as described above.

[0138] In the above embodiments, the data is divided into segments according to a preset time interval to obtain multiple segment data. Then, the segment data is determined to be a charging segment or a discharging segment based on the discharge and charging amounts during the time from the start time to the end time of the segment. After that, k typical charging segments and k typical discharging segments are selected as charging and discharging segments. By dividing the data into segments, this embodiment can reduce the computational workload of extracting statistical features for each segment, improve feature extraction efficiency, and increase the number of statistical features, thus providing support for model training.

[0139] In an exemplary embodiment, step 104 above, which involves performing feature distribution statistics on all types of packages for each statistical feature associated with each parameter characteristic to obtain a feature distribution, may include the following: calculating the probability density distribution of feature values ​​for each statistical feature of the charge and discharge segments of all battery packs corresponding to any type of package, and obtaining the feature distribution corresponding to each statistical feature; wherein, any type of package is a normal package or an abnormal package, and when the type of package is a normal package, the obtained feature distribution is the first feature distribution, and when the type of package is an abnormal package, the obtained feature distribution is the second feature distribution.

[0140] For any type of packet, the probability density distribution is calculated as follows:

[0141]

[0142] In a probability density distribution, any point y on the horizontal axis is an eigenvalue of a statistical feature f associated with a parametric property, and F(y) on the vertical axis is the probability density corresponding to any eigenvalue y on the horizontal axis. Let f be the eigenvalue of the i-th discharge segment or the i-th charging segment of any type of packet with respect to the statistical characteristic f. In the calculated probability density distribution, For each characteristic f, there is a statistical feature of the same type; K is the kernel function, h is the window parameter of the kernel function K; and d is any characteristic value. The dimension is d=1; M is the total number of all charge and discharge segments of the type packet for solving the probability density distribution, that is, M is the sum of the number of all charging segments and all discharging segments of the type packet for solving the probability density distribution.

[0143] It is worth noting that, for the statistical feature of the maximum single-cell voltage of a segment, a first characteristic distribution for normal packets and a second characteristic distribution for abnormal packets are obtained. Specifically, the maximum single-cell voltage of all charging segments and all discharging segments of all normal packets is obtained, forming the set of horizontal axis points of the first characteristic distribution with respect to the maximum single-cell voltage of the segment. Based on the probability density distribution formula F(y) above, the first characteristic distribution is obtained, where the first characteristic distribution for normal packets is P(y); the maximum single-cell voltage of all charging segments and all discharging segments of all abnormal packets is obtained, forming the set of horizontal points of the second characteristic distribution with respect to the maximum single-cell voltage of the segments: The second feature distribution is obtained according to the above probability density distribution formula F(y), where the first feature distribution of the abnormal packet is Q(y).

[0144] Understandably, for each type of statistical feature, a first feature distribution P(y) and a second feature distribution Q(y) can be obtained. That is, for the statistical feature of the maximum single-cell voltage of a segment, a first feature distribution P(y) for normal packets and a second feature distribution Q(y) for abnormal packets can be obtained. Similarly, for other types of statistical features, a first feature distribution for normal packets and a feature distribution for abnormal packets can be obtained.

[0145] In one exemplary embodiment, such as Figure 4 As shown, for step 105: determining statistically significant differences between normal and abnormal packets as training features for the training model based on the first and second feature distributions that correlate the same parameter characteristics, steps 401 to 403 may be included:

[0146] Step 401: Obtain the first feature distribution and the second feature distribution that are associated with the same parameter characteristics.

[0147] For example, for the parameter characteristic of voltage, a first feature distribution P(y) and a second feature distribution Q(y) are obtained for the statistical characteristic of the maximum single-unit voltage of the segment.

[0148] Step 402: Determine the feature difference value between normal packets and abnormal packets based on the intersection area of ​​the first feature distribution and the second feature distribution.

[0149] The feature difference value sdd between normal packets and abnormal packets for the same type of statistical features is:

[0150]

[0151] Where P(y) and Q(y) are the first and second characteristic distributions of the same type of statistical feature, respectively, and y is the horizontal axis value representing the statistical feature value in the characteristic distribution, where +∞ is positive infinity and -∞ is negative infinity. The area of ​​intersection between the first and second characteristic distributions is denoted as .

[0152] Taking the reciprocal of the intersection area yields the feature difference value sdd between the normal and abnormal packets.

[0153] Step 403: Based on the feature difference value and the preset difference value, determine whether the statistical features associated with the same parameter feature are statistical features with significant differences.

[0154] The feature difference value is compared with the preset difference value. If the feature difference value is greater than or equal to the preset difference value, the statistical feature associated with the same parameter is determined to be a statistical feature with significant difference. If the feature difference value is less than the preset difference value, the statistical feature associated with the same parameter is determined to be a statistical feature with significant difference.

[0155] In one embodiment, for all normal packets, a first feature distribution of statistical features of the type of maximum single-unit voltage of a fragment is obtained; for all abnormal packets, a second feature distribution of statistical features of the type of maximum single-unit voltage of a fragment is obtained; then, the feature difference value of the statistical feature of the type of maximum single-unit voltage of a fragment is calculated using the SDD calculation formula. If the feature difference value is greater than or equal to a preset difference value, then the statistical feature of maximum single-unit voltage of a fragment is used as the input feature of the training model. For other types of statistical features, the same method is used to calculate the feature difference value between normal packets and abnormal packets to determine whether other statistical features should be used as input to the training model.

[0156] Reference Figure 5a and Figure 5b In the figure, the horizontal axis represents the characteristic values ​​of the statistical features, and the vertical axis represents the probability density values ​​calculated based on the above expression for the probability density distribution.

[0157] In the above embodiments, feature selection based on the first feature distribution and the second feature distribution that are associated with the same parameter characteristics can yield statistical features with high discriminative power, thereby improving the quality of training features and thus improving the model quality, enabling the model to detect the health status of the battery pack more accurately.

[0158] Because abnormal packet data is more difficult to obtain than normal packet data, an imbalance can occur between the two types of data. Since the training model is based on both normal and abnormal packet data, this imbalance can affect the detection and recognition results of the training model. Therefore, it is necessary to construct abnormal packet data to address the lack of available data.

[0159] In one exemplary embodiment, such as Figure 6 As shown, the process of constructing abnormal packet data using normal packet data can include the following steps:

[0160] Step 501: Divide the data of each normal packet into multiple charge / discharge segments.

[0161] Step 502: Extract features from the charge and discharge segments of multiple normal packets to obtain feature vectors of multiple charge and discharge segments of each normal packet.

[0162] Step 503: Determine the probability density distribution of all feature vectors of the normal package based on the kernel function.

[0163] Step 504: Determine the inverse probability density distribution of the abnormal packet based on the probability density distribution.

[0164] Step 505: Sample the inverse probability density distribution to obtain abnormal packet data.

[0165] In the above embodiments, each normal packet data is divided into multiple charge / discharge segments; features are extracted from the charge / discharge segments of the multiple normal packets to obtain feature vectors for each normal packet's multiple charge / discharge segments; the probability density distribution of all feature vectors is determined based on a kernel function; the inverse probability density distribution belonging to the abnormal packet is determined based on the probability density distribution; and the inverse probability density distribution is sampled to obtain the abnormal packet data. This embodiment of the application fully utilizes the distribution information of the normal packet data to construct abnormal packet data, which can effectively address the problem of scarce abnormal samples, expand the number of abnormal samples, and improve the imbalance of training feature distribution.

[0166] It is worth noting that in step 501, the method for dividing each normal data packet into charge / discharge segments is the same as in step 102 above, and before performing step 501, the aforementioned data cleaning method can be used to clean each normal packet. In step 502, the extracted feature vectors include the following:

[0167] 1) Obtain the maximum voltage of all cells in the battery pack at each sampling time of each charge-discharge segment. Based on the maximum voltage of all sampling times of each charge-discharge segment, determine the maximum single cell voltage of each charge-discharge segment. The first feature is the maximum single cell voltage of the segment.

[0168] The formula for calculating the maximum single-unit voltage of the segment is the same as the calculation method described above.

[0169] 2) Obtain the minimum voltage of all cells in the battery pack at each sampling time of each charge-discharge segment. Based on the minimum voltage at all sampling times of each charge-discharge segment, determine the minimum single cell voltage of each charge-discharge segment. The second feature is the minimum single cell voltage of the segment.

[0170] The calculation formula for the minimum single-unit voltage of the segment is the same as the calculation method described above.

[0171] 3) Obtain the first mean or the first variance of the current value of the battery pack at all sampling times of each charge / discharge segment, and the third feature is the first mean or the first variance.

[0172] The expression for calculating the first mean is as follows:

[0173]

[0174] in, The first mean, Let be the current value of the battery pack at the i-th sampling time in the charge / discharge segment, and m be the number of sampling times in the charge / discharge segment.

[0175] The expression for calculating the first variance is as follows:

[0176]

[0177] in, The first variance, The first mean, Let be the current value of the battery pack at the i-th sampling time in the charge / discharge segment, and m be the number of sampling times in the charge / discharge segment.

[0178] 4) Obtain the second mean or second variance of the state of charge of the battery pack at all sampling times of each charge-discharge segment, and the fourth feature is the second mean or second variance.

[0179] The expression for calculating the second mean is as follows:

[0180]

[0181] in, The second mean, Let m be the state of charge value of the battery pack at the i-th sampling time in the charge / discharge segment, and m be the number of sampling times in the charge / discharge segment.

[0182] The expression for calculating the second variance is as follows:

[0183]

[0184] in, The second variance, The second mean, Let m be the state of charge value of the battery pack at the i-th sampling time in the charge / discharge segment, and m be the number of sampling times in the charge / discharge segment.

[0185] 5) Obtain the third mean or third difference of the temperature of the battery pack at all sampling times of each charge and discharge segment. The fifth feature is the third mean or third difference.

[0186] The formula for calculating the third mean is as follows:

[0187]

[0188] in, It is the third mean. Let be the temperature of the battery pack at the i-th sampling moment in the charge / discharge segment, and m be the number of sampling moments in the charge / discharge segment.

[0189] The expression for calculating the third difference is as follows:

[0190]

[0191] in, For third-party differences, It is the third mean. Let be the temperature of the battery pack at the i-th sampling moment in the charge / discharge segment, and m be the number of sampling moments in the charge / discharge segment.

[0192] The feature vector of each charge / discharge segment is composed of at least two of the first to fifth features.

[0193] For example, z i1 As the first feature, z i2 As the second feature, z i3 As the third feature, z i4 As the fourth feature, z i5 As the fifth feature, when the feature vector consists of the above five features, the feature vector z i It can be represented as (z) i1 ,z i2 ,z i3 ,z i4 ,z i5 ) T .

[0194] In one embodiment, regarding step 503, determining the probability density distribution of all feature vectors of the normal package based on the kernel function may include:

[0195] The expression for calculating the probability density distribution is as follows:

[0196]

[0197] In this probability density function, the eigenvector z j Let be the independent variable of the probability density distribution. For the eigenvector z j The probability density value of z; i Let z represent the feature vector of the i-th charging segment or the i-th discharging segment of any normal packet data, n be the total number of charging or discharging segments of all normal packets, K be the kernel function, and d be the dimension of the feature vector (feature vector z). i It consists of 5 features (d = 5), and h is the window parameter of the kernel function K, which controls the smoothness. The larger h is, the smoother it is.

[0198] The probability density distribution of all eigenvectors can be calculated using the above expression for probability density distribution.

[0199] In one embodiment, regarding step 504, determining the inverse probability density distribution belonging to the anomalous packet based on the probability density distribution may include the following:

[0200] The expression for calculating the inverse probability density distribution is as follows:

[0201]

[0202] in, For the eigenvector z j The probability density distribution, where n is the total number of charging or discharging segments of all normal packets. This represents the inverse probability density distribution of the abnormal packets.

[0203] In one embodiment, regarding step 505, the inverse probability density distribution is sampled to obtain abnormal packet data.

[0204] First, calculate the cumulative distribution function based on the inverse probability density distribution. The calculation expression is as follows:

[0205]

[0206] in, Let z be the inverse probability density distribution of the anomalous packets, and u be the feature vector z. j (z) j1 ,z j2 ,z j3 ,z j4 ,z j5 ) T Any value in the range.

[0207] Solve for the cumulative distribution function inverse function F -1 (u); then, multiple horizontal axis values ​​u are obtained by sampling from a standard uniform distribution; each horizontal axis value is substituted into the inverse function to obtain the anomaly packet data, i.e., w=F -1 (u), where w is the constructed exception package data.

[0208] In one exemplary embodiment, such as Figure 7 As shown, a method for detecting the health status of a battery pack is provided, which may include the following steps:

[0209] Step 601: Obtain multiple normal packet data and multiple abnormal packet data according to the data construction method in the above embodiments.

[0210] Step 602: Based on multiple normal packet data and multiple abnormal packet data, perform training feature filtering to determine the training features input into the training model.

[0211] Step 603: Train the model based on the training features to obtain the state recognition model.

[0212] Step 604: Use the state recognition model to perform state detection on the battery pack to be tested, and determine whether the battery pack is normal or abnormal.

[0213] In one embodiment, step 603, training the model using training features to obtain a state recognition model, may include the following:

[0214] The training model in this application can employ, but is not limited to, a one-class support vector machine (SVM). A state recognition model is obtained by training on a one-class SVM, using the highly discriminative training features selected in the above embodiments as input, and outputting the judgment result of the battery pack's health status, i.e., "normal" or "abnormal". One-class SVM is an unsupervised anomaly detection algorithm. Its core principle is to learn the distribution boundary of normal samples in a high-dimensional feature space, construct an optimal hyperplane, tightly wrapping the normal data, thereby identifying samples that significantly deviate from this distribution as abnormal.

[0215] In some embodiments, the training model can also be replaced with other machine learning algorithms suitable for unsupervised or semi-supervised anomaly detection, depending on the actual application scenario and data characteristics. These include, but are not limited to, K-Nearest Neighbors (KNN), Local Anomaly Factor (LOF), Isolation Forest, K-Means, DBSCAN, Support Vector Data Description (SVDD), or Autoencoder. All of these algorithms can achieve anomaly identification based on the distribution characteristics of normal samples or the distance / density relationship between data.

[0216] For example, KNN and LOF measure the degree of anomalies through local proximity, Isolation Forest detects outliers using random segmentation path lengths, K-Means and DBSCAN identify isolated clusters through clustering structures, SVDD learns the minimum hypersphere boundary, and autoencoders capture samples that deviate from normal patterns through reconstruction errors. These models can be flexibly selected according to computational resources, data scale, and real-time detection requirements, achieving equivalent or optimized detection of battery pack health status while maintaining consistency in input features.

[0217] During model training, normal packets are divided into training, validation, and test sets in a 6:2:2 ratio, and abnormal packets are also divided into training, validation, and test sets in a 6:2:2 ratio. Statistical features are extracted from the training set using the aforementioned statistical feature extraction method, and training features are selected using the aforementioned training feature selection method. The model is then trained based on these training features to obtain the state recognition model. Any existing training method can be used for model training.

[0218] Furthermore, the partitioning of training data can be further simplified by dividing only normal packets into training and testing sets, and only abnormal packets into training and testing sets, with a partitioning ratio of 8:2 or 7:3. With this simplified partitioning, a k-fold cross-validation strategy is used for model training and hyperparameter optimization: the training set is divided into k subsets. Each time, k-1 subsets are used for model training, and the remaining subset is used as the validation set to evaluate performance. This process is repeated k times, ensuring each subset participates in one validation iteration. This process yields k models with the same structure and hyperparameters but different specific parameters. The average and standard deviation of the validation performance of these k models are used to comprehensively evaluate model stability and hyperparameter effectiveness. After adjusting the hyperparameters, the above process is repeated, and the hyperparameters that produce the best average performance and standard deviation are selected as the optimal hyperparameters. Finally, the model is retrained using the entire training set and the selected optimal hyperparameters to obtain the final model. This method does not rely on an independent validation set, effectively improving the model's generalization ability and the reliability of parameter selection.

[0219] During the operation of the battery pack under test, battery pack data is acquired, the data is divided into charging and discharging segments, and features are extracted from each segment to obtain multiple statistical features corresponding to the battery pack under test. These statistical features are then input into a state recognition model, which performs state detection on the battery pack to determine whether it is functioning normally.

[0220] In the above embodiments, multiple normal packet data and multiple abnormal packet data are acquired according to the data construction method; training features are selected based on the multiple normal packet data and multiple abnormal packet data to determine the training features input into the training model; the model is trained using the training features to obtain a state recognition model; the state recognition model is used to perform state detection on the battery pack to be detected to determine whether the battery pack is normal or abnormal. Since the training features of the training model are selected based on the feature distribution of normal and abnormal packets, the quality of the training features is relatively good, which can improve the model quality and thus improve the accuracy of model detection.

[0221] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0222] Based on the same inventive concept, this application also provides a data construction apparatus for implementing the data construction method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more data construction apparatus embodiments provided below can be found in the limitations of the data construction method described above, and will not be repeated here.

[0223] In one exemplary embodiment, such as Figure 8 As shown, a data construction apparatus is provided, the apparatus comprising:

[0224] The segmentation module 701 is used to divide each normal packet data into multiple charge / discharge segments;

[0225] The feature extraction module 702 is used to extract features from the charging and discharging segments of multiple normal packets to obtain feature vectors of multiple charging and discharging segments of each normal packet.

[0226] The probability density determination module 703 is used to determine the probability density distribution of all feature vectors of the normal package based on the kernel function.

[0227] The distribution determination module 704 is used to determine the inverse probability density distribution belonging to the abnormal packet based on the probability density distribution;

[0228] The data acquisition module 705 is used to sample the inverse probability density distribution to obtain abnormal packet data.

[0229] It should be noted that each module in the data construction device can execute the above method embodiments, and their implementation principles and technical effects are similar, so they will not be described again here.

[0230] Each module in the aforementioned data construction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0231] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 9 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores data for constructing data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a data construction method.

[0232] Those skilled in the art will understand that Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0233] In one exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, executably including a memory of instructions that can be executed by a processor of an electronic device to perform the above-described method. Exemplarily, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0234] In one exemplary embodiment, a computer program product is also provided, which, when executed by a processor, can implement the above-described methods. The computer program product includes one or more computer instructions. When these computer instructions are loaded and executed on a computer, some or all of the above-described methods can be implemented, wholly or partially, according to the processes or functions described in the embodiments of this application.

[0235] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0236] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0237] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0238] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A data construction method, characterized in that, The method includes: The data from each normal battery pack is divided into multiple charge / discharge segments; Feature extraction is performed on the charging and discharging segments of multiple normal battery packs to obtain feature vectors of multiple charging and discharging segments of each normal battery pack; The probability density distribution of all feature vectors of the normal battery pack is determined based on the kernel function; Based on the probability density distribution, determine the inverse probability density distribution belonging to the abnormal battery pack; The inverse probability density distribution is sampled to obtain abnormal battery pack data; The step of sampling the inverse probability density distribution to obtain abnormal battery pack data includes: Calculate the cumulative distribution function based on the inverse probability density distribution; Find the inverse function of the cumulative distribution function; Samples are taken from a standard uniform distribution to obtain multiple horizontal axis values ​​for the standard uniform distribution; Substitute each horizontal axis value into the inverse function to obtain abnormal battery pack data; Wherein, the cumulative distribution function The calculation expression is: in, Let z be the inverse probability density distribution of the abnormal battery pack, and u be the eigenvector z. j (z) j1 ,z j2 ,z j3 ,z j4 ,z j5 ) T Any value in the range.

2. The method according to claim 1, characterized in that, The method further includes: Based on the battery voltage difference of multiple charging segments of each normal battery pack data, k typical charging segments of each normal battery pack data are determined. Based on the battery voltage difference of multiple discharge segments of each normal battery pack data, k typical discharge segments of each normal battery pack data are determined. The k typical charging segments and the k typical discharging segments are used as the charging and discharging segments for feature extraction, and each charging and discharging segment is either the charging segment or the discharging segment.

3. The method according to claim 1 or 2, characterized in that, The step of extracting features from the charge and discharge segments of the multiple normal battery packs to obtain feature vectors for the multiple charge and discharge segments of each normal battery pack includes: The maximum voltage of all cells in the battery pack at each sampling time of each charge-discharge segment is obtained. Based on the maximum voltage of all sampling times of each charge-discharge segment, the maximum single cell voltage of each charge-discharge segment is determined. The first feature is the maximum single cell voltage of the segment. The minimum voltage of all cells in the battery pack at each sampling time of each charge-discharge segment is obtained. Based on the minimum voltage at all sampling times of each charge-discharge segment, the minimum single cell voltage of each charge-discharge segment is determined. The second feature is the minimum single cell voltage of the segment. Obtain the first mean or the first variance of the current value of the battery pack at all sampling times of each charging and discharging segment, and the third feature is the first mean or the first variance. The second mean or second variance of the state of charge value of the battery pack at all sampling times of each charge-discharge segment is obtained, and the fourth feature is the second mean or the second variance; The third mean or third difference of the temperature of the battery pack at all sampling times of each charge-discharge segment is obtained, and the fifth feature is the third mean or the third difference. The feature vector of each of the charge / discharge segments is a feature vector composed of at least two of the first to the fifth features.

4. The method according to claim 1 or 2, characterized in that, The probability density distribution of all feature vectors of the normal battery pack is determined based on the kernel function, and the expression for calculating the probability density distribution is as follows: Wherein, the feature vector z j Let be the independent variable of the probability density distribution. For the eigenvector z j The probability density value of z; i Let represent the feature vector of the i-th charging segment or the i-th discharging segment of any normal battery pack data, n be the total number of charging or discharging segments of all normal battery packs, K be the kernel function, d be the dimension of the feature vector, and h be the window parameter of the kernel function K.

5. The method according to claim 4, characterized in that, The inverse probability density distribution for the abnormal battery pack is determined based on the probability density distribution, and the calculation expression for the inverse probability density distribution is as follows: in, For the eigenvector z j The probability density distribution is given by n, where n is the total number of charging or discharging segments of all normal battery packs. This represents the inverse probability density distribution of the abnormal battery pack.

6. The method according to claim 1 or 2, characterized in that, Before dividing the data of each normal battery pack into multiple charge / discharge segments, the data construction method further includes data cleaning for any type of pack, which can be a normal battery pack or an abnormal battery pack. The data cleaning process specifically includes: If the sample data in sample package A corresponding to each type of package meets the outlier condition, then sample package A of that type of package is cleared; if a sample package belonging to a type of package has missing sample data at time t, then the sample data of that sample package at time t is filled in using the sample data of the time adjacent to time t. Among them, sample package A is one of the following: battery pack current sample data, battery pack state of charge sample data, and cell voltage sample data corresponding to any type of package; the outlier condition is: the absolute difference between the sample data of two adjacent time points of sample package A is greater than a preset threshold.

7. The method according to claim 1 or 2, characterized in that, The charge / discharge segment is either a charging segment or a discharging segment. Multiple charge / discharge segments are divided from the data of each normal battery pack, including: According to a preset time interval, the data of each normal battery pack is divided into segments to obtain multiple data segments; If the discharge capacity of a data segment is greater than the charging capacity during the time from the start time of the segment to the end time of the segment, then the data segment is determined to be the discharge segment; If the discharge capacity of a data segment is less than the charging capacity during the time from the start time of the segment to the end time of the segment, then the data segment is determined to be the charging segment.

8. A method for detecting the health status of a battery pack, characterized in that, include: Acquire data from multiple normal battery packs and acquire data from multiple abnormal battery packs according to the data construction method described in any one of claims 1 to 7; Training features are selected based on multiple normal battery pack data and multiple abnormal battery pack data to determine the training features input into the training model. A state recognition model is obtained by training the model based on the training features. The state recognition model is used to detect the state of the battery pack under test and determine whether the battery pack is normal or abnormal.

9. A data construction apparatus, characterized in that, The device includes: The segmentation module is used to divide the data of each normal battery pack into multiple charge and discharge segments; The feature extraction module is used to extract features from the charging and discharging segments of multiple normal battery packs to obtain feature vectors of multiple charging and discharging segments of each normal battery pack. The probability density determination module is used to determine the probability density distribution of all feature vectors of the normal battery pack based on the kernel function; The distribution determination module is used to determine the inverse probability density distribution belonging to the abnormal battery pack based on the probability density distribution; The data acquisition module is used to sample the inverse probability density distribution to obtain abnormal battery pack data; Specifically, the data acquisition module is used to calculate the cumulative distribution function based on the inverse probability density distribution; solve for the inverse function of the cumulative distribution function; sample the standard uniform distribution to obtain multiple horizontal axis values ​​of the standard uniform distribution; and substitute each horizontal axis value into the inverse function to obtain abnormal battery pack data. The cumulative distribution function The calculation expression is: in, Let z be the inverse probability density distribution of the abnormal battery pack, and u be the eigenvector z. j (z) j1 ,z j2 ,z j3 ,z j4 ,z j5 ) T Any value in the range.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

Images