Battery pack fault detection method, battery pack, vehicle, device and equipment

Abstract

The embodiment of the application provides a battery pack fault detection method, a battery pack, a vehicle, an apparatus and equipment. It relates to the technical field of batteries. The method comprises the following steps: determining a plurality of data segments from charging time sequence voltage data; performing data filling on the plurality of data segments according to a preset time interval to construct a time sequence voltage matrix, wherein the time sequence voltage matrix comprises the voltage of each single battery cell of the battery pack to be detected; performing feature extraction on the time sequence voltage matrix through a multi-dimensional feature extraction method to obtain a high-dimensional feature set; and determining the fault detection result of each single battery cell according to the high-dimensional feature set and a fault classification model. The above scheme performs multi-dimensional feature recognition on the battery pack to be detected, captures early subtle differences from multiple dimensions and converts them into identifiable features, and then performs fault detection on the features through the fault classification model, so that the fault can be identified in advance, thereby improving the safety of the battery pack.

Description

Technical Field

[0001] This application relates to the field of battery technology, and in particular to a method for detecting faults in a battery pack, a battery pack, a vehicle, an apparatus, and equipment. Background Technology

[0002] In battery pack applications, the reliability of the battery pack directly impacts the safety of devices using it, such as electric vehicles and energy storage devices. Fault detection of the battery pack and the implementation of targeted strategies based on the detection results can improve safety.

[0003] In related technologies, fault detection of the battery pack is performed based on characteristic signals of the battery pack to determine whether a fault exists in the battery pack.

[0004] However, this method has a weak ability to identify minor faults in the early stages, and the fault detection is delayed, resulting in low battery pack safety. Summary of the Invention

[0005] This application provides a battery pack fault detection method, battery pack, vehicle, device, and equipment to improve the safety of the battery pack.

[0006] In a first aspect, embodiments of this application provide a fault detection method for a battery pack, comprising: responding to a fault identification request for the battery pack, determining charging time-series voltage data of the battery pack to be detected, and determining multiple data segments from the charging time-series voltage data, wherein the duration of each data segment is greater than or equal to a duration threshold; filling the multiple data segments with data according to a preset duration interval to construct a time-series voltage matrix, wherein the time-series voltage matrix includes the voltage of each individual cell of the battery pack to be detected; extracting features from the time-series voltage matrix using a multi-dimensional feature extraction method to obtain a high-dimensional feature set; determining a pre-trained fault classification model, and determining the fault detection result for each individual cell based on the high-dimensional feature set and the fault classification model.

[0007] Secondly, embodiments of this application provide a fault detection device for a battery pack, comprising: a screening module, configured to, in response to a fault identification request of the battery pack, determine the charging time-series voltage data of the battery pack to be detected, and determine multiple data segments from the charging time-series voltage data, wherein the duration of each data segment is greater than or equal to a duration threshold; a construction module, configured to fill the multiple data segments with data according to a preset duration interval to construct a time-series voltage matrix, wherein the time-series voltage matrix includes the voltage of each individual cell of the battery pack to be detected; an extraction module, configured to extract features from the time-series voltage matrix using a multi-dimensional feature extraction method to obtain a high-dimensional feature set; and a detection module, configured to, determine a pre-trained fault classification model, and determine the fault detection result of each individual cell based on the high-dimensional feature set and the fault classification model.

[0008] Thirdly, embodiments of this application provide a battery pack, the battery pack comprising multiple individual battery cells, wherein the fault detection result of each individual battery cell is no fault; the fault detection result is determined by the following method: in response to a fault identification request of the battery pack, the charging time-series voltage data of the battery pack to be tested is determined, and multiple data segments are determined from the charging time-series voltage data, the duration of each data segment being greater than or equal to a duration threshold; the multiple data segments are filled with data according to a preset duration interval to construct a time-series voltage matrix, the time-series voltage matrix including the voltage of each individual battery cell of the battery pack to be tested; features are extracted from the time-series voltage matrix through a multi-dimensional feature extraction method to obtain a high-dimensional feature set; a pre-trained fault classification model is determined, and the fault detection result of each individual battery cell is determined according to the high-dimensional feature set and the fault classification model.

[0009] Fourthly, embodiments of this application provide a vehicle that includes at least the battery pack described in the third aspect.

[0010] Fifthly, embodiments of this application provide a fault detection device for a battery pack, including: a memory and a processor;

[0011] The memory stores computer-executed instructions;

[0012] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0013] In a sixth aspect, embodiments of this application provide a non-volatile computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0014] In a seventh aspect, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0015] This application provides a battery pack fault detection method, battery pack, vehicle, device, and equipment. The method includes: responding to a battery pack fault identification request, determining the charging time-series voltage data of the battery pack to be detected, and determining multiple data segments from the charging time-series voltage data, each data segment having a duration greater than or equal to a duration threshold; filling the multiple data segments with data according to a preset duration interval to construct a time-series voltage matrix, the time-series voltage matrix including the voltage of each individual cell of the battery pack to be detected; extracting features from the time-series voltage matrix using a multi-dimensional feature extraction method to obtain a high-dimensional feature set; determining a pre-trained fault classification model; and determining the fault detection result for each individual cell based on the high-dimensional feature set and the fault classification model. This solution performs multi-dimensional feature identification on the battery pack to be detected, capturing early subtle differences from multiple dimensions and converting them into identifiable features. Then, by using a fault classification model to detect faults based on these features, faults can be identified in advance, thereby improving the safety of the battery pack. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] Figure 1 A schematic diagram illustrating an application scenario of a battery pack fault detection method provided in this application embodiment;

[0018] Figure 2 A schematic flowchart illustrating a battery pack fault detection method provided in an embodiment of this application;

[0019] Figure 3 A flowchart illustrating another battery pack fault detection method provided in this application embodiment;

[0020] Figure 4 A schematic diagram illustrating the distribution of statistical values ​​provided in the embodiments of this application;

[0021] Figure 5 A schematic diagram illustrating model-based fault prediction provided in an embodiment of this application;

[0022] Figure 6 A schematic diagram of the structure of a battery pack fault detection device provided in an embodiment of this application;

[0023] Figure 7A schematic diagram of the structure of another battery pack fault detection device provided in an embodiment of this application;

[0024] Figure 8 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0025] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0026] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0027] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0028] It should be noted that the phrase "at...time" in the embodiments of this application can refer to the instant at which a certain situation occurs, or to a period of time after the occurrence of a certain situation; the embodiments of this application do not specifically limit this. Furthermore, the display interface provided in the embodiments of this application is merely an example, and the display interface may include more or less content.

[0029] It should be noted that the fault detection method, battery pack, vehicle, device and equipment of this application can be used in the field of battery technology, or in any field other than batteries. The application field of the fault detection method, battery pack, vehicle, device and equipment of this application is not limited.

[0030] Figure 1This is a schematic diagram illustrating an application scenario of a battery pack fault detection method provided in this application embodiment. An example is given based on the illustrated scenario: fault detection is performed on a battery pack comprising multiple cells to identify the faulty cell within the battery pack.

[0031] For example, during the production process, battery packs may have latent faults such as abnormal internal resistance (e.g., lithium plating, metal particle foreign objects) inside the cells due to process defects, material impurities, or packaging problems. These latent faults are difficult to identify in conventional testing and may not immediately trigger obvious voltage or temperature anomalies in the early stages of battery pack operation.

[0032] As the battery pack is used cyclically, the internal resistance of faulty cells will be significantly higher than that of normal cells, causing a deviation between the voltage change rate of the faulty cells and the overall voltage consistency of the battery pack. If these high-risk cells are not detected and isolated in time, it may trigger local thermal runaway, leading to battery pack safety issues.

[0033] In related technologies, the Battery Management System (BMS) detects faults by using voltage outlier algorithms and negative temperature coefficient thermistors (NTCs) to detect temperature anomalies.

[0034] Among them, the voltage outlier algorithm monitors the voltage difference of each individual cell in the battery pack. When the voltage of a certain individual cell deviates from the overall average voltage by more than a preset threshold, it is determined to be an abnormal cell.

[0035] Among them, the NTC temperature anomaly detection uses a temperature sensor to monitor the temperature rise and combines it with a temperature threshold to detect faults.

[0036] However, the voltage outlier algorithm and NTC temperature anomaly detection have low sensitivity to early faults. They can only trigger warnings when the voltage difference or temperature difference reaches the threshold, at which point obvious faults have already occurred. Fault detection is lagging, resulting in low battery pack safety.

[0037] The method for detecting battery pack faults provided in this application aims to solve the aforementioned technical problems in related technologies.

[0038] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0039] Figure 2This is a flowchart illustrating a battery pack fault detection method provided in an embodiment of this application. The method includes the following steps:

[0040] S201. In response to a fault identification request for the battery pack, determine the charging timing voltage data of the battery pack to be tested, and determine multiple data segments from the charging timing voltage data, wherein the duration of each data segment is greater than or equal to a duration threshold.

[0041] The battery pack to be tested includes multiple individual battery cells.

[0042] For example, charging timing voltage data is collected through the BMS of the battery pack under test. The charging timing voltage data is real-time data of the battery pack under test during charging, and may include multiple timestamps, the voltage of each individual cell at each timestamp, and operating parameters (such as charging current, charging voltage, etc.).

[0043] For example, charging timing voltage data can be used for more accurate fault detection compared to timing voltage data under other operating conditions.

[0044] Based on scenario examples, battery pack operating conditions can be categorized into charging, discharging, and resting conditions. Discharging is a condition characterized by fluctuating loads. During discharge, the load changes dynamically, resulting in large current fluctuations. The battery pack voltage is significantly affected by the load than by its own faults, and fault characteristics are often masked by load noise. In resting conditions, the battery pack voltage tends to balance, reflecting only the static voltage difference and failing to capture dynamic voltage change rate characteristics. Charging is a constant current and constant voltage controlled condition. The BMS maintains a constant charging current, and the internal resistance differences of the cells are directly converted into voltage differences, leading to higher accuracy in fault detection.

[0045] For example, the duration threshold is the minimum charging duration that can reflect the voltage variation pattern of the battery cell.

[0046] With scenario examples, filtering data segments can remove voltage data corresponding to abnormal charging conditions (such as charging interruption, current surges exceeding preset thresholds, and abnormal NTC temperature); and voltage data during the constant current charging phase can also be filtered. This ensures that the charging conditions within each data segment are stable, the data is continuous, and there is no significant noise interference.

[0047] S202. Data is filled into multiple data segments according to a preset time interval to construct a time-series voltage matrix, which includes the voltage of each individual cell of the battery pack to be tested.

[0048] For example, multiple data segments are collected at a preset collection frequency. When the collection frequency is low, data gaps may occur. Data is filled in according to preset time intervals to enrich the data content, thereby more accurately capturing early subtle faults. When multiple data segments are collected at different frequencies, data filling at preset time intervals can unify the time dimension and avoid detection errors introduced by the time dimension.

[0049] To illustrate with a scenario example, if the collection period for charging timing voltage data is 10 minutes and the preset time interval is 100 ms, the collection frequency is low and the data density is low. Data filling can increase the data density and thus enrich the data content.

[0050] Optionally, the timing voltage matrix includes time, individual cell identifier, and voltage.

[0051] With the help of a scenario example, each element (i, j) in the time-series voltage matrix can represent the voltage of the single cell with cell identifier j at time i.

[0052] Based on the above implementation methods, discrete charging timing voltage data is transformed into a standardized and structured timing voltage matrix, thereby unifying the data dimensions, eliminating interference caused by missing data, and improving the accuracy of fault detection.

[0053] S203. By using a multi-dimensional feature extraction method, features are extracted from the time-series voltage matrix to obtain a high-dimensional feature set.

[0054] For example, by using a multi-dimensional feature extraction method, features of voltage in the time-series voltage matrix are mined and extracted from multiple different perspectives related to single cell failures.

[0055] Specifically, the time-series voltage matrix is ​​a structured set of the time-series voltages of each individual cell. The multi-dimensional feature extraction method is based on this matrix, simultaneously capturing the time-series voltage variation characteristics of each individual cell and the voltage difference characteristics of each individual cell relative to other cells in the battery pack. Through quantification of these features, subtle performance differences that exist in the early stages of cell development and cannot be identified by conventional single-feature extraction methods (such as subtle voltage fluctuations caused by small changes in internal resistance, subtle deviations in voltage rise rate, etc.) are transformed into distinguishable feature parameters.

[0056] For example, all quantized feature parameters are integrated to form a high-dimensional feature set. Each feature parameter in the high-dimensional feature set is associated with the corresponding individual cell, which can comprehensively and accurately characterize the charging voltage characteristics of each individual cell, thereby effectively identifying early subtle faults.

[0057] S204. Determine the pre-trained fault classification model, and determine the fault detection result for each individual cell based on the high-dimensional feature set and the fault classification model.

[0058] For example, the pre-trained fault classification model analyzes and identifies the feature parameters corresponding to each individual cell, and combines the feature difference patterns between normal cells and faulty cells learned during the model training process to judge the working status of each individual cell, and finally outputs the fault detection result corresponding to each individual cell.

[0059] With the help of scenario examples, the fault classification model is trained based on a large number of feature parameters. It can capture subtle changes in feature parameters. Compared with voltage outlier algorithms and NTC temperature anomaly detection, it can accurately identify individual cells in the early stages of faults.

[0060] Optionally, the fault detection results include the identification of the faulty individual cell. Based on the identification, the BMS can implement targeted strategies for the faulty individual cell in the early stages of the fault, thereby improving the safety of the battery pack under test.

[0061] The battery pack fault detection method provided in this application, in response to a battery pack fault identification request, determines the charging time-series voltage data of the battery pack to be detected, and identifies multiple data segments from the charging time-series voltage data, each data segment having a duration greater than or equal to a duration threshold; the multiple data segments are filled with data according to a preset duration interval to construct a time-series voltage matrix, the time-series voltage matrix including the voltage of each individual cell of the battery pack to be detected; features are extracted from the time-series voltage matrix through a multi-dimensional feature extraction method to obtain a high-dimensional feature set; a pre-trained fault classification model is determined, and the fault detection result of each individual cell is determined based on the high-dimensional feature set and the fault classification model. This solution performs multi-dimensional feature identification on the battery pack to be detected, capturing early subtle differences from multiple dimensions and converting them into identifiable features. Then, by using a fault classification model to detect faults based on these features, faults can be identified in advance, thereby improving the safety of the battery pack.

[0062] Based on any of the above embodiments, the following, in conjunction with Figure 3 The detailed process of fault detection for battery packs is explained.

[0063] Figure 3 This is a flowchart illustrating another battery pack fault detection method provided in an embodiment of this application. Figure 3 As shown, the method includes:

[0064] S301. In response to a fault identification request for the battery pack, determine the charging timing voltage data of the battery pack to be tested, and determine multiple data segments from the charging timing voltage data, wherein the duration of each data segment is greater than or equal to a duration threshold.

[0065] One feasible implementation method is to determine multiple data segments by: determining the voltage change pattern of the battery pack to be tested; determining the corresponding duration threshold based on the voltage change pattern and the identification sensitivity; and selecting multiple data segments from the charging timing voltage data based on the duration threshold, wherein the data of each data segment is continuous and the duration of each data segment is greater than or equal to the duration threshold.

[0066] The fault identification request includes identification sensitivity.

[0067] For example, the sensitivity setting characterizes the precision with which the fault detection process captures early-stage faults. A higher sensitivity setting allows the detection process to capture more subtle performance differences in individual cells; a lower sensitivity setting allows for the identification of more obvious faults, prioritizing detection efficiency.

[0068] For example, voltage variation patterns are used to clarify the voltage variation characteristics of the battery pack under test. During periods of gentle voltage changes and small fluctuations, longer data segments are needed to capture subtle fault differences. During periods of drastic voltage changes and large fluctuations, shorter data segments are sufficient to reflect fault characteristics, providing a basis for determining subsequent duration thresholds.

[0069] Optionally, the charging timing voltage data can be analyzed using a BMS to determine the voltage variation pattern.

[0070] To illustrate with scenario examples, during the constant current charging phase, the cell voltage increases at a uniform rate over time, with gradual changes and minimal fluctuations. During the constant voltage charging phase, the voltage tends to stabilize, and the rate of change decreases significantly. Brief voltage fluctuations may occur at the beginning and / or end of the charging process.

[0071] For example, by combining the voltage change pattern and the recognition sensitivity, a time threshold that meets the requirements is determined to balance detection accuracy and detection efficiency.

[0072] For example, each data segment obtained through screening is free of missing or interrupted data (data fragments caused by signal interference or charging interruptions are removed), ensuring that the data can completely reflect the voltage change characteristics of a single battery cell within that time period, and avoiding incomplete data from affecting subsequent feature extraction and fault detection accuracy. The duration of each data segment is greater than or equal to a dynamically determined duration threshold, ensuring that the amount of data in each segment is sufficient to support subsequent multi-dimensional feature extraction and meet the fault detection requirements at the corresponding recognition sensitivity.

[0073] In this feasible implementation, the specified data segment duration ensures sufficient data volume, supporting multi-dimensional features to identify subtle changes in individual cell voltage from different perspectives. Data continuity ensures that voltage differences corresponding to early, minor faults are not masked, allowing multi-dimensional features to accurately identify subtle performance differences that conventional methods cannot detect. Therefore, by combining voltage change patterns and identification sensitivity, faults can be identified earlier, thereby improving battery pack safety.

[0074] S302. A data segment sequence is obtained by combining the voltage change pattern and the recognition sensitivity according to the time sequence. The data segment sequence includes the voltage of each individual cell.

[0075] For example, the segments are arranged in chronological order, with the segments corresponding to the earlier charging periods listed first and the segments corresponding to the later charging periods listed last, thus completely restoring the timing logic of the entire charging process of the battery pack under test.

[0076] Based on the above implementation method, discrete data segments are integrated into a coherent charging timing data set through time-series arrangement, ensuring that subsequent filling and matrix construction can fully preserve the timing variation characteristics of the cell voltage. These timing variation characteristics are key to capturing subtle differences in early-stage cell faults, thereby enabling early fault identification.

[0077] S303. Using a preset time interval as the time granularity, fill the data segment sequence with interpolation or data fitting methods to obtain the time-series voltage matrix.

[0078] For example, a preset time interval is used to unify the data sampling time granularity. The voltage data of the entire data segment sequence is processed into a uniform voltage sampling point at each preset time interval, ensuring that the padded data is evenly distributed and dimensionally uniform in the time dimension. This resolves the subtle differences that may exist in the original sampling intervals of different data segments (such as fluctuations that occur during BMS acquisition) and ensures the standardization of the time dimension of the time-series voltage matrix.

[0079] With the example of the scenario, although the selected data segments meet the continuity requirement, there may be a small amount of missing voltage data within each data segment due to signal interference. By using interpolation or data fitting methods, these missing points can be accurately filled based on adjacent effective voltage data, ensuring that each unified voltage sampling point has corresponding individual cell voltage data, and avoiding missing data from masking the subtle voltage differences in the early stages of the cell.

[0080] Optionally, the timing voltage matrix is ​​represented by Table 1:

[0081] Table 1

[0082]

[0083] Referring to Table 1, the voltage of cell A at time 19:23:47 is 3.596, and the other data in Table 1 are similar. A uniform voltage sampling point is 1 second, meaning one voltage is recorded every 1 second. It can be seen that the voltages of different cells in the time-series voltage matrix are close at the same time; therefore, detection methods that rely on identifying signals with significant characteristics cannot detect early, subtle faults.

[0084] Based on the above implementation methods, by constructing a time-series voltage matrix that is sequential, complete in data, and unified in dimensions, it is ensured that during subsequent multi-dimensional feature extraction, subtle early voltage changes in individual cells can be accurately captured from multiple perspectives, such as time dimension and group difference dimension, thereby enabling early fault identification.

[0085] S304. By using a multi-dimensional feature extraction method, feature extraction is performed on the time-series voltage matrix to obtain multiple extracted features.

[0086] Among them, the multi-dimensional feature extraction methods include at least one of the following: statistical feature extraction, weighted feature extraction, difference feature extraction, statistical test feature extraction, and frequency domain feature extraction.

[0087] One feasible implementation method is to extract features through statistical feature extraction, including: resampling the time-series voltage matrix according to multiple sampling frequencies to obtain multiple time-series voltage sub-matrices; for each time-series voltage sub-matrix, extracting the voltage local statistical features of each individual cell, which include at least one of the following: mean, median, first quartile, third quartile, maximum value, minimum value, standard deviation, skewness, and kurtosis.

[0088] For example, the local statistical features of voltage are the results of statistical feature extraction.

[0089] For example, multiple sampling frequencies correspond to different time granularities. By resampling the time-series voltage matrix based on these sampling frequencies, the original time-series voltage matrix can be aggregated according to different time granularities to generate multiple time-series voltage sub-matrices with different time resolutions.

[0090] To illustrate with a scenario example, the original time-series voltage matrix has a time granularity of 100ms. If three sampling frequencies of 1s, 5s, and 10s are set, the voltage data of the original matrix are aggregated at time intervals of 1s, 5s, and 10s respectively, resulting in three time-series voltage sub-matrices. The only difference between the time-series voltage sub-matrices and the time-series voltage matrix is ​​that the number of voltage sampling points decreases as the sampling frequency decreases.

[0091] Optionally, the local statistical characteristics of voltage are shown in Table 2:

[0092] Table 2

[0093]

[0094] Referring to Table 2, the average voltage of cell A at multiple moments is 3.595. The maximum voltage of cell B at multiple moments is 3.597. Other local voltage statistical characteristics are similar. These different local voltage statistical characteristics allow for the assessment of individual cell voltage variations from various perspectives.

[0095] Optionally, the skewness can be calculated using the following formula:

[0096]

[0097] Where E[⋅] represents the calculated mean, X represents the variable value, i.e., voltage, μ represents the variable average, and σ represents the standard deviation.

[0098] Alternatively, kurtosis can be calculated using the following formula:

[0099]

[0100] In this feasible implementation, short-time granularity easily captures instantaneous, minute voltage fluctuations, while long-time granularity highlights slow voltage shifts. By generating sub-matrices through multiple sampling frequencies, these subtle changes can be identified simultaneously at different time granularities, preventing minor differences at a single time granularity from being masked by noise. This allows for early fault identification, thereby improving battery pack safety.

[0101] One feasible implementation method is to extract features through weighted feature extraction, including: calculating the weight value of the voltage of each individual cell at each time step in the time-series voltage matrix relative to the total voltage of multiple individual cells, and constructing a time-series weight matrix; resampling the time-series weight matrix according to multiple sampling frequencies to obtain multiple weight sub-matrices; and extracting the weighted local statistical features of each individual cell for each weight sub-matrix.

[0102] For example, the local statistical features of the weights are the extraction results of the weight features.

[0103] For example, based on the time-series voltage matrix, with each time moment as an independent calculation unit, the weight value of each individual cell at each time moment is calculated as the weight value of its voltage value relative to the sum of the voltages of multiple individual cells at that time moment.

[0104] Optionally, the weight value can be represented by the following formula:

[0105]

[0106] Where t represents time and i represents the identifier of a single battery cell. Let represent the voltage of cell i at time t, and n represent the total number of cells. This represents the voltage weight value of a single cell i.

[0107] It should be noted that the resampling method for the time-series weight matrix is ​​the same as that for the time-series voltage matrix, and will not be repeated here.

[0108] Optional, the weighted local statistical features include, but are not limited to, at least one of the following: mean, median, first quartile, third quartile, maximum value, minimum value, standard deviation, skewness, and kurtosis.

[0109] Optionally, the local statistical characteristics of the weights are shown in Table 3:

[0110] Table 3

[0111]

[0112] Referring to Table 3, the voltage weight of cell A at time 19:23:45 is 0.009525. The average weight of cell B at multiple times is 0.009532. The maximum weight of cell C at multiple times is 0.009531. Other local statistical characteristics of the weights are similar. By using different local statistical characteristics of the weights, the changes in the weights of individual cells can be evaluated from different dimensions.

[0113] In this feasible implementation, the voltage changes caused by early, minute faults in a single battery cell are extremely small and difficult to detect in the voltage dimension. However, by calculating weight values, the voltage of the single cell can be correlated with the sum of the group voltages, transforming minute voltage deviations into proportional changes in relative weight. This relative proportion representation method can amplify the voltage differences caused by early, minute faults, making voltage differences that would otherwise be undetectable by conventional methods stand out in the weight dimension. This allows for early fault identification, thereby improving the safety of the battery pack.

[0114] One feasible implementation involves feature extraction via a difference feature extraction method, including: calculating the global voltage statistical parameters of multiple individual cells at each time step in the time-series voltage matrix, where the global voltage statistical parameters include at least one of the following: mean, median, first quartile, third quartile, maximum value, and minimum value; performing difference operations between each voltage submatrix and the corresponding global voltage statistical parameters at the sampling frequency to obtain a voltage difference matrix; and extracting voltage difference features from the voltage difference matrix, where the voltage difference features include at least one of the following: mean, median, first quartile, third quartile, and maximum value. The calculation of weighted global statistical parameters for multiple individual cells at each time step in the time-series voltage matrix includes at least one of the following: mean, median, first quartile, third quartile, maximum, and minimum. For each weighted submatrix, the difference is calculated with the corresponding weighted global statistical parameter at the sampling frequency to obtain a weighted difference matrix. Weighted difference features are extracted from the weighted difference matrix, including at least one of the following: mean, median, first quartile, third quartile, maximum, minimum, standard deviation, skewness, and kurtosis.

[0115] For example, voltage difference features and weighted difference features are the extraction results of difference feature extraction.

[0116] For example, the global voltage statistics parameter represents the overall voltage statistics of multiple individual battery cells. The global weight statistics parameter represents the overall weight statistics of multiple individual battery cells.

[0117] Optionally, the global voltage statistics are shown in Table 4:

[0118] Table 4

[0119]

[0120] Referring to Table 4, at time 19:23:46, the average voltage of multiple individual cells is 3.592. At time 23:02:29, the maximum voltage of multiple individual cells is 4.184. The same applies to other global voltage statistical parameters. By constructing a group distribution benchmark for the voltage of individual cells at each time point using global voltage statistical parameters, the voltage distribution characteristics can be comprehensively quantified.

[0121] Optionally, the global statistical parameters of the weights are shown in Table 5:

[0122] Table 5

[0123]

[0124] Referring to Table 5, at time 19:23:46, the minimum weight value of multiple individual cells is 0.009522. At time 23:02:30, the median weight value of multiple individual cells is 0.009527. The same applies to other global statistical parameters of the weights. By constructing a group distribution benchmark for the weights of individual cells at each time point using global statistical parameters of the weights, the weight distribution characteristics can be comprehensively quantified.

[0125] For example, for each voltage submatrix, the voltage global statistical parameters with the same sampling frequency as that voltage submatrix (i.e., the voltage global statistical parameters are resampled at the corresponding sampling frequency to ensure their time scale is perfectly aligned with the voltage submatrix) are used to perform interpolation. The interpolation is performed on a per-cell basis to obtain the difference for each per-cell at each time step. A voltage difference matrix is ​​then constructed based on the differences for each per-cell.

[0126] Optionally, for each global statistical parameter, the corresponding voltage difference matrix can be calculated separately.

[0127] Optionally, the formula for calculating the difference can be: individual cell voltage difference = cell voltage value in the voltage submatrix - global voltage statistics parameters at the same moment.

[0128] Optionally, the voltage difference characteristics are shown in Table 6:

[0129] Table 6

[0130]

[0131] Referring to Table 6, the voltage difference of cell A at time 19:23:45 is 0.000448. The average voltage difference of cell B at multiple times is 0.007263. The weighted median of cell C at multiple times is 0.000096. Other voltage difference characteristics are similar. Different voltage difference characteristics can be used to evaluate the temporal characteristics of the voltage difference of a single cell from different dimensions.

[0132] Optionally, the voltage difference can be calculated based on different sampling frequencies. For example, if the sampling frequency is 10s, a voltage difference value is taken every 10s from the voltage difference data of the same single cell in the voltage difference matrix, and feature calculations (such as mean, median, first quartile, third quartile, maximum value, minimum value, standard deviation, skewness, and kurtosis) are performed on the taken voltage difference values ​​to obtain the voltage difference characteristics corresponding to the sampling frequency of 10s.

[0133] Similarly, a weight difference matrix with the same dimensions as the corresponding weight submatrix is ​​generated. Each element in the weight difference matrix represents the degree of deviation of the weight value of a single cell from the overall weight benchmark value at a certain moment.

[0134] In this feasible implementation, the relative deviation is directly quantified into a difference value through difference calculation, which is equivalent to linearly amplifying early subtle differences. This allows for early identification of faults, thereby improving the safety of the battery pack.

[0135] One feasible implementation method involves feature extraction through statistical testing, including: extracting the voltage sequence of each individual cell for each voltage sub-matrix; performing a two-sample distribution comparison test on the voltage sequence and the corresponding global voltage statistical parameters to obtain the first test feature; extracting the weight sequence of each individual cell for each weight sub-matrix; and performing a two-sample distribution comparison test on the weight sequence and the corresponding global weight statistical parameters to obtain the second test feature.

[0136] For example, the first and second test features are the results of statistical test feature extraction.

[0137] For example, by taking each individual cell as the extraction unit in the voltage submatrix, the voltage values ​​of that individual cell at all times in the voltage submatrix are arranged in chronological order to obtain a voltage sequence. The voltage sequence can represent the voltage time-series variation pattern of an individual cell at the current time granularity.

[0138] For example, the voltage sequence of each individual cell is taken as the first sample, and the global statistical parameter sequence of voltage at the same sampling frequency is taken as the second sample. A two-sample distribution comparison test is performed, and the degree of distribution difference between the two samples is quantified by the test results. The generated test statistic, significance index and other parameters are the first test features of the individual cell at the corresponding sampling frequency.

[0139] Alternatively, the two-sample distribution comparison test can be the Mann-Whitney test and / or the Kolmogorov-Smirnov (KS) test.

[0140] Optionally, the Mann-Whitney test can be specifically performed by comparing two samples. and The samples are merged into a single whole sample, which contains a total of One observation value.

[0141] Sort all observations in ascending order and assign them a rank. If there are duplicate values, take the average rank.

[0142] The candidate statistic is calculated using the following formula:

[0143]

[0144]

[0145] in, and This indicates the statistical quantity to be selected.

[0146] Pick and The minimum value in the range is used as the statistic.

[0147] Calculate the mean:

[0148]

[0149] in, This represents the mean.

[0150] Calculate the standard deviation:

[0151]

[0152] in, Indicates standard deviation, This indicates the number of repetitions of the k-th node (with the same value), and g represents the number of groups of nodes.

[0153] Calculate the Z-score; the absolute value of the Z-score indicates the degree of difference in the distribution.

[0154]

[0155] Optionally, the KS test method can specifically be:

[0156] For the sample After sorting, it is ≤ ≤...≤ Empirical Cumulative Distribution Function (ECDF) Defined as:

[0157]

[0158] Where I represents the indicative function, when If x ≤ x, I = 1; otherwise, I = 0.

[0159] Calculate the test statistic D:

[0160]

[0161] in: This represents the empirical cumulative distribution function of the first group of samples. Let represent the empirical cumulative distribution function of the second group of samples, and sup denote the supremum (maximum value).

[0162] When the sample size is large ( , (>100), the critical value can be calculated using an approximate formula. :

[0163]

[0164] The calculation uses an approximate formula for the p-value:

[0165]

[0166] The D-statistic represents the degree of difference in the overall distribution patterns of two samples. A larger D-statistic indicates a greater deviation between the voltage or weight sequence of a single battery cell and the cumulative distribution curve of the global statistical parameter sequence, suggesting a more anomalous distribution characteristics of the single battery cell. (Critical value) The D statistic is used to determine whether it is statistically significant. The p-value represents the probability that the distributions of two sequences are indistinguishable. The smaller the p-value, the higher the significance of the distribution difference, and the greater the possibility of early failure in a single battery cell.

[0167] It should be noted that the extraction method for the second test feature is the same as that for the first test feature, and will not be repeated here.

[0168] Below, in conjunction with Figure 4 Explain the distribution of statistical values.

[0169] Figure 4 This is a schematic diagram illustrating the distribution of statistical values ​​provided in an embodiment of this application. For example... Figure 4 As shown, this includes the voltage value sequence of one individual battery cell and multiple global voltage statistical parameters at the same sampling frequency. The distribution patterns and numerical concentration ranges of the voltage sequence of the individual battery cell and the sequences of each global voltage statistical parameter differ only slightly; these differences cannot be directly quantified by a single numerical value. Early subtle differences between individual battery cells and the group benchmark can be identified through statistical feature extraction.

[0170] In this feasible implementation, by performing a two-sample distribution comparison test on the voltage sequence and the global statistical parameters of voltage, and the weight sequence and the global statistical parameters of weight, the subtle early differences between individual cells and the group benchmark are captured from the data distribution level. This makes up for the shortcomings of the global statistical features of voltage, the local statistical features of weight, and the difference features, which can only characterize the magnitude and dispersion of values. It enables the accurate identification of subtle deviations in the distribution pattern caused by early faults of individual cells, and can identify faults in advance, thereby improving the safety of the battery pack.

[0171] One feasible implementation involves feature extraction via frequency domain feature extraction, including: performing frequency domain transformation on the voltage sequence of each individual cell in the time-series voltage matrix to obtain a frequency domain voltage matrix; resampling the frequency domain voltage matrix according to multiple sampling frequencies to obtain multiple frequency domain voltage sub-matrices; calculating the global statistical parameters of the frequency domain voltage of multiple individual cells corresponding to each time step in each frequency domain voltage sub-matrix; and extracting the local statistical features of the frequency domain voltage of each individual cell for each frequency domain voltage sub-matrix, wherein the local statistical features of the frequency domain voltage include at least one of the following: mean, median, first quartile, third quartile, and maximum value. The following parameters are used to calculate the voltage difference matrix: mean, median, first quartile, third quartile, maximum, minimum, standard deviation, skewness, and kurtosis. For each frequency domain voltage submatrix, the difference is calculated with the corresponding frequency domain global statistical parameters at the sampling frequency to obtain the frequency domain voltage difference matrix. Frequency domain voltage difference features are extracted from the frequency domain voltage difference matrix, including at least one of the following: mean, median, first quartile, third quartile, maximum, minimum, standard deviation, skewness, and kurtosis. For each frequency domain voltage submatrix, the frequency domain voltage sequence of each individual cell is extracted. A two-sample distribution comparison test is performed on the frequency domain voltage sequence and the corresponding frequency domain voltage global statistical parameters to obtain the third test feature.

[0172] For example, the frequency domain voltage local statistical features, frequency domain voltage difference features, and third test features are the extraction results of frequency domain feature extraction.

[0173] Optionally, frequency domain transformation can be performed via Fourier transform.

[0174] For example, based on the time-series voltage matrix, frequency domain transformation is performed on the voltage sequence corresponding to each individual cell, converting the voltage-time variation signal in the time domain into a voltage amplitude-frequency variation signal in the frequency domain. Following a structure where frequency points are rows and individual cell identifiers are columns, a frequency domain voltage matrix matching the dimensions of the time-series voltage matrix is ​​constructed. Each element in the matrix represents the voltage amplitude at the corresponding frequency point and for the corresponding individual cell.

[0175] It should be noted that the resampling method is the same as the above scheme, and will not be repeated here.

[0176] For example, for each frequency domain voltage submatrix, each frequency point (corresponding to each moment in the time domain) is used as an independent calculation unit to calculate the frequency domain voltage amplitude of multiple individual cells at that frequency point, and calculate the global statistical parameters of the frequency domain voltage.

[0177] Optional, the global statistical parameters of the frequency domain voltage include, but are not limited to, at least one of the following: mean, median, first quartile, third quartile, maximum value, and minimum value.

[0178] For example, for each frequency domain voltage sub-matrix, each individual cell is taken as an independent extraction unit, and local statistical features of the frequency domain voltage are extracted for all frequency domain voltage amplitudes of that individual cell in the frequency domain voltage sub-matrix.

[0179] Optionally, the local statistical characteristics of the frequency domain voltage include, but are not limited to, at least one of the following: mean, median, first quartile, third quartile, maximum value, minimum value, standard deviation, skewness, and kurtosis.

[0180] For example, for each frequency domain voltage submatrix, a difference operation is performed using global statistical parameters of the frequency domain voltage at the same sampling frequency as that submatrix. A frequency domain voltage difference matrix is ​​constructed based on the difference operation results. The frequency domain voltage difference characteristics are then calculated based on the frequency domain voltage difference matrix.

[0181] For example, the frequency domain voltage sequence of each individual cell is used as the first sample, and the corresponding frequency domain voltage global statistical parameter sequence at the same frequency point is used as the second sample. A two-sample distribution comparison test is performed to obtain the third test feature.

[0182] In this feasible implementation, voltage changes caused by early minor faults in a single cell are easily masked by charging noise in the time domain, but will manifest as abnormal amplitudes of specific frequency components in the frequency domain. By converting the voltage signal to the frequency dimension through frequency domain transformation, it is possible to accurately capture early subtle differences in the frequency dimension that cannot be identified by time domain features, thereby identifying faults in advance and improving the safety of the battery pack.

[0183] S305. Group the multiple extracted features to obtain multiple initial feature groups. Each initial feature group includes the extracted features corresponding to the same single cell.

[0184] For example, the extracted features are grouped based on individual cells, and all features belonging to the same individual cell are grouped together. Each group corresponds to one individual cell, and finally, multiple initial feature groups are obtained, which are the same as the number of individual cells in the battery pack.

[0185] S306. According to the preset feature order, the extracted features in each initial feature group are spliced ​​together to obtain multiple spliced ​​feature groups.

[0186] For example, the preset feature order is a predefined feature arrangement rule that is uniformly applied to all initial feature groups, and the order corresponds one-to-one with the feature extraction method.

[0187] For example, the splicing operation is performed independently for each initial feature group. According to the preset feature order, all extracted feature values ​​in the initial feature group are concatenated to form a one-dimensional feature vector. After splicing, each initial feature group yields a corresponding spliced ​​feature group (i.e., the standardized feature vector of a single cell).

[0188] S307. Arrange multiple spliced ​​feature groups to construct a high-dimensional feature set.

[0189] For example, the arrangement operation is based on the numbering order of individual battery cells. The resulting multiple spliced ​​feature groups are arranged in the order of battery cell numbering and combined to form a two-dimensional feature matrix. This two-dimensional feature matrix is ​​the high-dimensional feature set.

[0190] For example, the rows of the high-dimensional feature set correspond to individual battery cells, and the columns correspond to the feature dimensions spliced ​​in a preset order. Each row is a complete feature vector of an individual battery cell, and each column is the feature value of all individual battery cells under the same feature dimension. The structure is regular and can be directly input into a pre-trained fault classification model for early fault identification of battery cells.

[0191] S308. Determine the pre-trained fault classification model, and determine the fault detection result for each individual cell based on the high-dimensional feature set and the fault classification model.

[0192] One feasible implementation method is to determine the pre-trained fault classification model by: determining the target vehicle identifier of the vehicle to which the battery pack to be tested is applied; determining the mapping relationship between the vehicle identifier and the fault classification model; and determining the pre-trained fault classification model based on the target vehicle identifier and the mapping relationship.

[0193] For example, the battery to be predicted is installed in a specific vehicle, and the target vehicle identifier is a unique identifier of the vehicle in which the battery to be predicted is installed, in order to distinguish it from other vehicles.

[0194] For example, the mapping relationship is a pre-established one-to-one correspondence. The mapping relationship is established based on the fact that the battery pack parameters and operating conditions of different vehicles are different, and the corresponding fault classification model must also be adapted to its operating conditions in order to ensure the accuracy of fault identification.

[0195] For example, the process of determining the pre-trained fault classification model corresponding to the battery pack to be detected is as follows: using the target vehicle identifier as an index, query the corresponding fault classification model from the mapping relationship.

[0196] With the help of scenario examples, it can be seen that the operating conditions and parameters of battery packs vary significantly among different vehicles. A single fault classification model cannot be adapted to all vehicles and is prone to errors in fault feature identification, missed detections, and false detections due to mismatched operating conditions.

[0197] In this feasible implementation, by mapping the vehicle identifier to the corresponding fault classification model, it is ensured that the model used for the battery pack under test is trained based on samples of the vehicle's battery pack, accurately matching its operating conditions. This effectively avoids identification errors caused by differences in operating conditions, improving the accuracy and adaptability of fault detection.

[0198] A feasible implementation method for training a fault classification model includes: acquiring sample charging time-series voltage data and corresponding fault labels; extracting features from the sample charging time-series voltage data using a multi-dimensional feature extraction method to obtain a training feature set; grouping the training feature set according to the fault labels to obtain fault sample groups and normal sample groups; selecting key features from the fault sample groups and normal sample groups whose differences exceed a preset threshold using a statistical significance test method; and training a binary classification model based on a gradient boosting tree algorithm using the key features and corresponding fault labels as samples to obtain the fault classification model.

[0199] For example, a large amount of time-series voltage data of battery packs in different health states are collected during the charging process, including normal cell samples and faulty cell samples.

[0200] For example, features are extracted from the sample charging time-series voltage data using the same multi-dimensional feature extraction method as when constructing the high-dimensional feature set, to obtain a training feature set. Based on the fault label, the training feature set is then binary-grouped to obtain a fault sample group and a normal sample group.

[0201] For example, by using a statistical significance test method applicable to numerical features, the difference test is performed on the feature values ​​of the same feature dimension in the fault sample group and the normal sample group to quantify the degree of difference of each feature dimension between the fault sample group and the normal sample group, and the significance of the difference, i.e., the p-value, is calculated.

[0202] By comparing the significance of the differences with the prediction threshold, features with differences exceeding the threshold are retained, while redundant features with no significant differences (such as features affected by operating noise or features that cannot be distinguished as fault samples) are removed. The resulting feature set is the key feature.

[0203] Optionally, during the model inference process, key features are extracted from the high-dimensional feature set and then input into the fault classification model for fault detection.

[0204] For example, gradient boosting trees (such as GBDT and XGBoostLightGBM) are used to construct a binary classification model. The selected key features are used as model input, and the corresponding fault labels are used as model output. Cross-validation is then used to train the model to obtain the fault classification model.

[0205] Below, in conjunction with Figure 5This paper explains model-based fault prediction.

[0206] Figure 5 This is a schematic diagram illustrating model-based fault prediction provided in an embodiment of this application. Figure 5 As shown, basic data is prepared for model training by collecting charging time-series voltage samples containing normal and early-faulty cells as the data source for model training. The collected sample charging time-series voltage data is preprocessed to remove outliers, missing values, and invalid sampling points to ensure the integrity and accuracy of the sample data. Feature extraction is performed using a multi-dimensional feature extraction method, and a training feature set is constructed. Based on the actual health status detection results of individual cells, each individual cell sample in the training feature set is labeled with a fault tag. Key features with differences exceeding a preset threshold are selected from the faulty sample group and the normal sample group. Redundant noise features are removed through significance testing, retaining the key features that can distinguish between faulty and normal samples, resulting in a key feature set for model training. A binary classification model based on the gradient boosting tree algorithm is trained using the key feature set to obtain a fault classification model. The performance of the fault classification model is evaluated, and the model with the strongest generalization ability and the highest fault identification accuracy is selected as the optimal fault classification model. Based on the known operating conditions of faulty battery packs and the model's predicted probabilities, customized screening conditions are established: When applying the optimal fault classification model to actual cell fault detection, historical operating conditions of known faulty battery packs and the model's output fault prediction probabilities are combined to formulate fault judgment screening conditions suitable for actual application scenarios. Using the optimal fault classification model, high-dimensional feature sets obtained after extracting the same features from the actual collected cell charging time-series voltage data are used to predict faults, outputting fault prediction results and identifying individual cells with early-stage faults. For each faulty individual cell, subsequent fault handling operations are performed to eliminate the fault.

[0207] In this feasible implementation, key features with significant differences are screened out through statistical significance testing. This can reduce the computational complexity of the model, improve training efficiency, and avoid overfitting caused by the model learning invalid noise features, thereby improving the accuracy of model inference.

[0208] This application provides a battery pack comprising multiple individual battery cells, wherein the fault detection result of each individual battery cell is no fault. The fault detection result is determined as follows: in response to a fault identification request of the battery pack, the charging time-series voltage data of the battery pack to be tested is determined, and multiple data segments are determined from the charging time-series voltage data, the duration of each data segment being greater than or equal to a duration threshold; the multiple data segments are filled with data according to a preset duration interval to construct a time-series voltage matrix, the time-series voltage matrix including the voltage of each individual battery cell of the battery pack to be tested; features are extracted from the time-series voltage matrix through a multi-dimensional feature extraction method to obtain a high-dimensional feature set; a pre-trained fault classification model is determined, and the fault detection result of each individual battery cell is determined based on the high-dimensional feature set and the fault classification model.

[0209] Based on the above implementation method, multi-dimensional feature recognition is performed on the battery pack to be tested. Early subtle differences are captured from multiple dimensions and transformed into identifiable features. Then, by using a fault classification model to detect faults in the features, faults can be identified in advance, thereby improving the safety of the battery pack.

[0210] It should be noted that the process of determining the fault detection result of the battery pack shown in the embodiments of this application can be implemented by the technical solution shown in the above method embodiments. The implementation principle and beneficial effects are similar, and will not be repeated here.

[0211] This application provides a vehicle that includes at least the aforementioned battery pack.

[0212] Based on the above implementation method, the battery pack installed in the vehicle is detected and verified through multi-dimensional feature recognition technology. It can capture early subtle differences from multiple dimensions and transform them into identifiable features, which can identify faults in advance, thereby improving vehicle safety.

[0213] Figure 6 This is a schematic diagram of a battery pack fault detection device provided in an embodiment of this application. Figure 6 As shown, the fault detection device 60 for the battery pack may include: a screening module 61, a construction module 62, an extraction module 63, and a detection module 64.

[0214] The filtering module 61 is used to respond to the fault identification request of the battery pack, determine the charging timing voltage data of the battery pack to be tested, and determine multiple data segments from the charging timing voltage data, wherein the duration of each data segment is greater than or equal to the duration threshold.

[0215] The construction module 62 is used to fill multiple data segments with data according to a preset time interval to construct a time-series voltage matrix, which includes the voltage of each individual cell of the battery pack to be tested.

[0216] The extraction module 63 is used to extract features from the time-series voltage matrix through a multi-dimensional feature extraction method to obtain a high-dimensional feature set.

[0217] The detection module 64 is used to determine the pre-trained fault classification model and, based on the high-dimensional feature set and the fault classification model, determine the fault detection result for each individual cell.

[0218] Optionally, the filtering module 61 can perform... Figure 2 S201 in the embodiment.

[0219] Optionally, builder module 62 can be executed. Figure 2 S202 in the embodiment.

[0220] Optionally, extraction module 63 can be executed. Figure 2 S203 in the embodiment.

[0221] Optionally, the detection module 64 can perform... Figure 2 S204 in the embodiment.

[0222] Based on the above implementation method, multi-dimensional feature recognition is performed on the battery pack to be tested. Early subtle differences are captured from multiple dimensions and transformed into identifiable features. Then, by using a fault classification model to detect faults in the features, faults can be identified in advance, thereby improving the safety of the battery pack.

[0223] It should be noted that the battery pack fault detection device shown in the embodiments of this application can execute the technical solution shown in the above method embodiments, and its implementation principle and beneficial effects are similar, so they will not be described again here.

[0224] In one possible implementation, the fault identification request includes identification sensitivity; the screening module 61 is specifically used for:

[0225] Determine the voltage variation pattern of the battery pack under test;

[0226] Based on the voltage change pattern and recognition sensitivity, determine the corresponding duration threshold;

[0227] Based on the duration threshold, multiple data segments are selected from the charging timing voltage data. The data in each segment is continuous, and the duration of each segment is greater than or equal to the duration threshold.

[0228] In one possible implementation, the extraction module 63 is specifically used for:

[0229] The time-series voltage matrix is ​​subjected to feature extraction using a multi-dimensional feature extraction method, resulting in multiple extracted features. The multi-dimensional feature extraction method includes at least one of the following: statistical feature extraction, weighted feature extraction, difference feature extraction, statistical test feature extraction, and frequency domain feature extraction.

[0230] Multiple extracted features are grouped to obtain multiple initial feature groups, each of which includes extracted features corresponding to the same single battery cell;

[0231] According to the preset feature order, the extracted features in each initial feature group are concatenated to obtain multiple concatenated feature groups;

[0232] Multiple concatenated feature groups are arranged to construct a high-dimensional feature set.

[0233] In one possible implementation, the extraction module 63 is specifically used for:

[0234] The time-series voltage matrix is ​​resampled based on multiple sampling frequencies to obtain multiple time-series voltage sub-matrices;

[0235] For each time-series voltage submatrix, local voltage statistical features of each individual cell are extracted. The local voltage statistical features include at least one of the following: mean, median, first quartile, third quartile, maximum value, minimum value, standard deviation, skewness, and kurtosis.

[0236] In one possible implementation, the extraction module 63 is specifically used for:

[0237] In the time-series voltage matrix, the weight value of the voltage of each individual cell at each time point relative to the total voltage of multiple individual cells is calculated, and a time-series weight matrix is ​​constructed.

[0238] The time-series weight matrix is ​​resampled based on multiple sampling frequencies to obtain multiple weight submatrices;

[0239] For each weight submatrix, extract the local statistical features of the weights for each individual battery cell.

[0240] In one possible implementation, the extraction module 63 is specifically used for:

[0241] In the time-series voltage matrix, calculate the global statistical parameters of the voltage of multiple individual cells at each time step. The global statistical parameters of voltage include at least one of the following: mean, median, first quartile, third quartile, maximum value, and minimum value.

[0242] For each voltage submatrix, a difference operation is performed with the corresponding global voltage statistical parameters at the sampling frequency to obtain the voltage difference matrix.

[0243] Extract the voltage difference features of each individual cell from the voltage difference matrix. The voltage difference features include at least one of the following: mean, median, first quartile, third quartile, maximum value, minimum value, standard deviation, skewness, and kurtosis.

[0244] In the calculation of the time-series voltage matrix, the weighted global statistical parameters of multiple individual cells at each time step include at least one of the following: mean, median, first quartile, third quartile, maximum value, and minimum value.

[0245] For each weight submatrix, a difference operation is performed between it and the corresponding global statistical parameters of the weights at the sampling frequency to obtain the weight difference matrix.

[0246] Extract the weight difference features of each individual cell from the weight difference matrix. The weight difference features include at least one of the following: mean, median, first quartile, third quartile, maximum value, minimum value, standard deviation, skewness, and kurtosis.

[0247] In one possible implementation, the extraction module 63 is specifically used for:

[0248] For each voltage submatrix, extract the voltage sequence of each individual battery cell;

[0249] The first test feature is obtained by comparing the voltage sequence and the corresponding global voltage statistics with two samples;

[0250] For each weight submatrix, extract the weight sequence for each individual battery cell;

[0251] The second test feature is obtained by comparing the weight sequence and the corresponding global statistical parameters of the weights using a two-sample distribution.

[0252] In one possible implementation, the extraction module 63 is specifically used for:

[0253] Perform frequency domain transformation on the voltage sequence of each individual cell in the time-series voltage matrix to obtain the frequency domain voltage matrix;

[0254] The frequency domain voltage matrix is ​​resampled based on multiple sampling frequencies to obtain multiple frequency domain voltage sub-matrices;

[0255] Calculate the global statistical parameters of the frequency domain voltage of multiple individual cells at each time step in each frequency domain voltage submatrix;

[0256] For each frequency domain voltage sub-matrix, extract the local statistical features of the frequency domain voltage for each individual cell. The local statistical features of the frequency domain voltage include at least one of the following: mean, median, first quartile, third quartile, maximum value, minimum value, standard deviation, skewness, and kurtosis.

[0257] For each frequency domain voltage submatrix, the difference operation is performed with the corresponding frequency domain global statistical parameters at the sampling frequency to obtain the frequency domain voltage difference matrix.

[0258] Extract frequency domain voltage difference features from the frequency domain voltage difference matrix. The frequency domain voltage difference features include at least one of the following: mean, median, first quartile, third quartile, maximum value, minimum value, standard deviation, skewness, and kurtosis.

[0259] For each frequency domain voltage submatrix, extract the frequency domain voltage sequence of each individual battery cell;

[0260] The third test feature is obtained by comparing the frequency domain voltage sequence and the corresponding global statistical parameters of the frequency domain voltage with two samples.

[0261] Figure 7 This is a schematic diagram of the structure of another battery pack fault detection device provided in an embodiment of this application. Figure 6 Based on the illustrated embodiments, as Figure 7 As shown, the fault detection device 60 for the battery pack also includes: an execution module 65 and a training module 66.

[0262] Execution module 65 is used for:

[0263] Multiple data segments are arranged in chronological order to obtain a data segment sequence, which includes the voltage of each individual battery cell.

[0264] Using a preset time interval as the time granularity, the data segment sequence is filled by interpolation or data fitting to obtain the time-series voltage matrix.

[0265] Training module 66 is used for:

[0266] Acquire sample charging timing voltage data and corresponding fault tags;

[0267] By using a multi-dimensional feature extraction method, features are extracted from the sample charging time-series voltage data to obtain a training feature set;

[0268] Based on the fault labels, the training feature set is grouped into fault sample groups and normal sample groups;

[0269] By using statistical significance testing, key features that differ from the faulty sample group and the normal sample group by a preset threshold are selected.

[0270] Using key features and corresponding fault labels as samples, a binary classification model based on the gradient boosting tree algorithm is trained to obtain a fault classification model.

[0271] Figure 8This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application, such as... Figure 8 As shown, the electronic device includes:

[0272] The electronic device includes a processor 291 and a memory 292; it may also include a communication interface 293 and a bus 294. The processor 291, memory 292, and communication interface 293 can communicate with each other via the bus 294. The communication interface 293 can be used for information transmission. The processor 291 can invoke logical instructions stored in the memory 292 to execute the methods of the above embodiments.

[0273] Furthermore, the logic instructions in the aforementioned memory 292 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0274] The memory 292, as a non-volatile computer-readable storage medium, can be used to store software programs and computer-executable programs, such as program instructions / modules corresponding to the methods in the embodiments of this application. The processor 291 executes functional applications and data processing by running the software programs, instructions, and modules stored in the memory 292, that is, it implements the methods in the above-described method embodiments.

[0275] The memory 292 may include a program storage area and a data storage area. The program storage area may store the operating system and application programs required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 292 may include high-speed random access memory and may also include non-volatile memory.

[0276] Based on the above implementation method, multi-dimensional feature recognition is performed on the battery pack to be tested. Early subtle differences are captured from multiple dimensions and transformed into identifiable features. Then, by using a fault classification model to detect faults in the features, faults can be identified in advance, thereby improving the safety of the battery pack.

[0277] This application provides a non-volatile computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method as described in the foregoing embodiments.

[0278] Based on the above implementation method, multi-dimensional feature recognition is performed on the battery pack to be tested. Early subtle differences are captured from multiple dimensions and transformed into identifiable features. Then, by using a fault classification model to detect faults in the features, faults can be identified in advance, thereby improving the safety of the battery pack.

[0279] This application provides a computer program product, including a computer program that, when executed by a processor, implements the method as described in the foregoing embodiments.

[0280] Based on the above implementation method, multi-dimensional feature recognition is performed on the battery pack to be tested. Early subtle differences are captured from multiple dimensions and transformed into identifiable features. Then, by using a fault classification model to detect faults in the features, faults can be identified in advance, thereby improving the safety of the battery pack.

[0281] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0282] It should be further noted that although the steps in the flowchart 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; they can be executed in other orders. Moreover, at least some steps in the flowchart may include multiple sub-steps or multiple stages, which do not necessarily complete at the same time but can be executed at different times. The execution order of these sub-steps or stages is also not necessarily sequential but can be alternated or carried out in turn with other steps or at least some of the sub-steps or stages of other steps.

[0283] It should be understood that the above-described device embodiments are merely illustrative, and the device of this application can also be implemented in other ways. For example, the division of units / modules in the above embodiments is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple units, modules, or components may be combined, or integrated into another system, or some features may be ignored or not executed.

[0284] Furthermore, unless otherwise specified, the functional units / modules in the various embodiments of this application can be integrated into one unit / module, or each unit / module can exist physically separately, or two or more units / modules can be integrated together. The integrated units / modules described above can be implemented in hardware or as software program modules.

[0285] When integrated units / modules are implemented in hardware, the hardware can be digital circuits, analog circuits, etc. The physical implementation of the hardware structure includes, but is not limited to, transistors, memristors, etc. The processor can be any suitable hardware processor, such as CPU, GPU, FPGA, DSP, and ASIC. The storage unit can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc.

[0286] If the integrated unit / module is implemented as a software program module and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to related technologies, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0287] In the above embodiments, the descriptions of each embodiment have their own emphasis. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments. The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0288] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the claims.

[0289] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A method for detecting faults in a battery pack, characterized in that, include: In response to a battery pack fault identification request, the charging timing voltage data of the battery pack to be tested is determined, and multiple data segments are determined from the charging timing voltage data, the duration of each data segment being greater than or equal to a duration threshold. The multiple data segments are filled with data according to a preset time interval to construct a time-series voltage matrix, which includes the voltage of each individual cell of the battery pack to be tested. By using a multi-dimensional feature extraction method, features are extracted from the time-series voltage matrix to obtain a high-dimensional feature set; A pre-trained fault classification model is determined, and the fault detection result for each individual battery cell is determined based on the high-dimensional feature set and the fault classification model.

2. The method according to claim 1, characterized in that, Determine the pre-trained fault classification model, including: Determine the target vehicle identifier of the vehicle to which the battery pack to be tested is applied; Determine the mapping relationship between vehicle identification and fault classification model; The pre-trained fault classification model is determined based on the target vehicle identifier and the mapping relationship.

3. The method according to claim 1, characterized in that, The fault identification request includes identification sensitivity; it determines multiple data segments from the charging timing voltage data, including: Determine the voltage variation pattern of the battery pack under test; Based on the voltage change pattern and the recognition sensitivity, determine the corresponding duration threshold; Based on the duration threshold, the plurality of data segments are selected from the charging timing voltage data, wherein the data of each data segment is continuous and the duration of each data segment is greater than or equal to the duration threshold.

4. The method according to claim 3, characterized in that, Data is filled into the multiple data segments according to a preset time interval to construct a time-series voltage matrix, including: The multiple data segments are arranged in chronological order to obtain a data segment sequence, wherein the data segment sequence includes the voltage of each individual battery cell; Using the preset time interval as the time granularity, the data segment sequence is filled by interpolation or data fitting to obtain the time-series voltage matrix.

5. The method according to claim 1, characterized in that, By employing a multi-dimensional feature extraction method, features are extracted from the time-series voltage matrix to obtain a high-dimensional feature set, including: The time-series voltage matrix is ​​subjected to feature extraction using a multi-dimensional feature extraction method to obtain multiple extracted features. The multi-dimensional feature extraction method includes at least one of the following: statistical feature extraction, weighted feature extraction, difference feature extraction, statistical test feature extraction, and frequency domain feature extraction. The extracted features are grouped to obtain multiple initial feature groups, each initial feature group including the extracted features corresponding to the same single battery cell; According to the preset feature order, the extracted features in each initial feature group are concatenated to obtain multiple concatenated feature groups; The multiple spliced ​​feature groups are arranged to construct the high-dimensional feature set.

6. The method according to claim 5, characterized in that, The time-series voltage matrix is ​​subjected to feature extraction using statistical feature extraction methods, including: The time-series voltage matrix is ​​resampled according to multiple sampling frequencies to obtain multiple time-series voltage sub-matrices; For each time-series voltage sub-matrix, local voltage statistical features of each individual cell are extracted. The local voltage statistical features include at least one of the following: mean, median, first quartile, third quartile, maximum value, minimum value, standard deviation, skewness, and kurtosis.

7. The method according to claim 6, characterized in that, The time-series voltage matrix is ​​subjected to feature extraction using a weighted feature extraction method, including: Calculate the weight value of the voltage of each individual cell at each time point in the time-series voltage matrix relative to the total voltage of multiple individual cells, and construct a time-series weight matrix; Based on the multiple sampling frequencies, the time-series weight matrix is ​​resampled to obtain multiple weight submatrices; For each weighted submatrix, extract the weighted local statistical features of each individual battery cell.

8. The method according to claim 7, characterized in that, Feature extraction is performed on the time-series voltage matrix using a difference feature extraction method, including: Calculate the global statistical parameters of the voltage of multiple individual cells corresponding to each time step in the time-series voltage matrix. The global statistical parameters of the voltage include at least one of the following: mean, median, first quartile, third quartile, maximum value, and minimum value. For each voltage submatrix, a difference operation is performed with the corresponding global voltage statistical parameters at the sampling frequency to obtain the voltage difference matrix. The voltage difference features of each individual cell are extracted from the voltage difference matrix. The voltage difference features include at least one of the following: mean, median, first quartile, third quartile, maximum value, minimum value, standard deviation, skewness, and kurtosis. Calculate the weighted global statistical parameters of the plurality of individual cells corresponding to each time step in the time-series voltage matrix. The weighted global statistical parameters include at least one of the following: mean, median, first quartile, third quartile, maximum value, and minimum value. For each weight submatrix, a difference operation is performed between it and the corresponding global statistical parameters of the weights at the sampling frequency to obtain the weight difference matrix. The weight difference features of each individual cell are extracted from the weight difference matrix. The weight difference features include at least one of the following: mean, median, first quartile, third quartile, maximum value, minimum value, standard deviation, skewness, and kurtosis.

9. The method according to claim 8, characterized in that, The time-series voltage matrix is ​​feature extracted using a statistical test feature extraction method, including: For each voltage sub-matrix, extract the voltage sequence of each individual battery cell; The first test feature is obtained by performing a two-sample distribution comparison test on the voltage sequence and the corresponding global voltage statistical parameters. For each weight submatrix, extract the weight sequence for each individual battery cell; The second test feature is obtained by performing a two-sample distribution comparison test on the weight sequence and the corresponding global statistical parameters of the weight.

10. The method according to claim 5, characterized in that, Feature extraction is performed on the time-series voltage matrix using a frequency domain feature extraction method, including: For the voltage sequence of each individual cell in the time-series voltage matrix, a frequency domain transformation is performed to obtain a frequency domain voltage matrix; The frequency domain voltage matrix is ​​resampled according to multiple sampling frequencies to obtain multiple frequency domain voltage sub-matrices; Calculate the global statistical parameters of the frequency domain voltage of multiple individual cells at each time step in each frequency domain voltage submatrix; For each frequency domain voltage sub-matrix, local statistical features of the frequency domain voltage of each individual cell are extracted. The local statistical features of the frequency domain voltage include at least one of the following: mean, median, first quartile, third quartile, maximum value, minimum value, standard deviation, skewness, and kurtosis. For each frequency domain voltage submatrix, the difference operation is performed with the corresponding frequency domain global statistical parameters at the sampling frequency to obtain the frequency domain voltage difference matrix. Frequency domain voltage difference features are extracted from the frequency domain voltage difference matrix. The frequency domain voltage difference features include at least one of the following: mean, median, first quartile, third quartile, maximum value, minimum value, standard deviation, skewness, and kurtosis. For each frequency domain voltage submatrix, extract the frequency domain voltage sequence for each individual battery cell; A third test feature is obtained by performing a two-sample distribution comparison test on the frequency domain voltage sequence and the corresponding frequency domain voltage global statistical parameters.

11. The method according to any one of claims 1-10, characterized in that, The training method for the fault classification model includes: Acquire sample charging timing voltage data and corresponding fault tags; By using a multi-dimensional feature extraction method, features are extracted from the sample charging time-series voltage data to obtain a training feature set; Based on the fault labels, the training feature set is grouped to obtain fault sample groups and normal sample groups; By using statistical significance testing, key features that differ from the faulty sample group and the normal sample group by a preset threshold are selected. Using the key features and corresponding fault labels as samples, the binary classification model constructed based on the gradient boosting tree algorithm is trained to obtain the fault classification model.

12. A battery pack, characterized in that, The battery pack includes multiple individual cells, wherein the fault detection result of each individual cell is no fault; The fault detection result is determined in the following way: In response to a battery pack fault identification request, the charging timing voltage data of the battery pack to be tested is determined, and multiple data segments are determined from the charging timing voltage data, the duration of each data segment being greater than or equal to a duration threshold. The multiple data segments are filled with data according to a preset time interval to construct a time-series voltage matrix, which includes the voltage of each individual cell of the battery pack to be tested. By using a multi-dimensional feature extraction method, features are extracted from the time-series voltage matrix to obtain a high-dimensional feature set; A pre-trained fault classification model is determined, and the fault detection result for each individual battery cell is determined based on the high-dimensional feature set and the fault classification model.

13. A vehicle, characterized in that, It includes at least the battery pack as described in claim 12.

14. A fault detection device for a battery pack, characterized in that, include: The filtering module is used to respond to the fault identification request of the battery pack, determine the charging timing voltage data of the battery pack to be detected, and determine multiple data segments from the charging timing voltage data, wherein the duration of each data segment is greater than or equal to a duration threshold. A construction module is used to fill the multiple data segments with data according to a preset time interval to construct a time-series voltage matrix, wherein the time-series voltage matrix includes the voltage of each individual cell of the battery pack to be tested; The extraction module is used to extract features from the time-series voltage matrix using a multi-dimensional feature extraction method to obtain a high-dimensional feature set; The detection module is used to determine the pre-trained fault classification model and, based on the high-dimensional feature set and the fault classification model, determine the fault detection result for each individual battery cell.

15. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1-11.

16. A non-volatile computer-readable storage medium, characterized in that, The non-volatile computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-11.

Images