A method and system for detecting leakage of a power battery

Abstract

This invention relates to the field of battery monitoring technology, specifically to a method and system for detecting leakage in power batteries. It includes the following steps: S1, acquiring the voltage sequence and insulation resistance of individual battery cells in a battery management system in real time; S2, standardizing the voltage sequence of the cleaned battery cells and extracting the minimum voltage value and standard deviation according to a preset time window to construct a two-dimensional feature vector; S3, inputting the two-dimensional feature vector into a multivariate exponentially weighted moving average model to calculate the cumulative anomaly, and smoothing the cumulative anomaly using a moving average. If the smoothed cumulative anomaly exceeds a first preset threshold, a level one alarm is triggered; S4, after the level one alarm is triggered, locating the lowest voltage point within the alarm period, and extracting a local voltage window sequence centered on that point, calculating the Z-score of the lowest voltage point and the proportion of the lowest voltage value in the total data, wherein the Z-score is used to measure the degree of local deviation.

Description

Technical Field

[0001] This invention relates to the field of battery monitoring technology, and specifically to a method and system for detecting leakage in power batteries. Background Technology

[0002] With the rapid development of the new energy vehicle industry, the power battery, as the core energy source of electric vehicles, directly affects the overall operational reliability of the vehicle and the life and property safety of passengers. Among the various failure modes of power batteries, electrolyte leakage is a common and serious type of failure, usually caused by factors such as poor battery sealing structure, external mechanical impact damage, aging of cell materials, or manufacturing defects. Electrolyte leakage not only leads to the loss of active materials inside the battery, causing accelerated capacity decay and abnormally increased internal resistance, but also releases corrosive and toxic gases. These gases can erode the electrical connectors and insulation structures inside the battery module, leading to serious safety accidents such as battery short circuits, thermal runaway, and even fires. Therefore, timely and accurate detection of power battery leakage is of paramount importance for ensuring the safe operation of electric vehicles.

[0003] Currently, battery leakage detection mainly relies on the following technical methods. The first is offline detection, which involves pressurizing or vacuuming the battery pack using specialized airtightness testing equipment during vehicle maintenance or troubleshooting, or manually disassembling the battery pack for visual inspection of the cells and connectors. While this method can accurately identify leakage faults, the detection process must be performed with the vehicle stopped, and the disassembly is destructive, damaging the battery pack's sealing structure. It is costly, time-consuming, and cannot achieve real-time online monitoring of leakage faults. The second method is online monitoring based on a single parameter. Some battery management systems collect the insulation resistance value of the battery pack in real time. When the insulation resistance value drops sharply below a safe threshold, an insulation fault alarm is triggered, and maintenance personnel use this to determine if leakage is possible. This method achieves online monitoring, but its judgment relies on a single factor. The decrease in insulation resistance value can be caused by various factors besides leakage, such as changes in environmental humidity, condensation accumulation, and electrical system interference, leading to a high false alarm rate. The third method is based on voltage anomaly detection. It determines whether there is leakage by monitoring the abnormal drop in the voltage of individual battery cells. However, voltage drop may also be caused by battery imbalance, loose connection or fluctuations in normal operating conditions. It also has a high false alarm rate and is not sensitive to the slight voltage changes in the early stage of leakage, making it difficult to achieve early warning.

[0004] A comprehensive analysis of existing technologies reveals the following key shortcomings. First, the issues of detection lag and destructiveness are significant. Traditional offline detection methods cannot achieve real-time online monitoring; by the time a fault is detected, irreversible battery damage has often already occurred. Furthermore, disassembly and inspection damage the original sealing structure of the battery pack, increasing repair costs and the risk of secondary faults. Second, there is a lack of early warning capabilities. The characteristic signals of early-stage leakage faults, such as slight drops in cell voltage and slow decay of insulation resistance, are easily masked by fluctuations in the normal operating conditions of the vehicle. Existing single-parameter monitoring methods struggle to effectively identify these weak signals from background noise, failing to trigger warnings before significant electrolyte leakage, thus missing the optimal time for intervention. Third, a high false alarm rate leads to insufficient diagnostic reliability. The threshold judgments for insulation resistance or voltage by single-parameter monitoring strategies are easily affected by external factors such as environmental humidity, temperature changes, and electrical system fluctuations. Frequent false alarms not only increase the workload of maintenance personnel but also reduce the reliability of the fault diagnosis system, potentially causing genuine leakage faults to be overlooked. Fourth, the lack of multivariate collaborative analysis capability is a significant drawback. Leakage faults are inherently complex processes involving the coupling of multiple physical quantities, manifesting as both voltage anomalies and insulation degradation, with specific temporal correlations between the two. Traditional methods have failed to establish effective multivariate dynamic correlation analysis models, making it difficult to accurately distinguish leakage faults from other types of faults such as individual cell imbalances or loose connections. This compromises the accuracy and reliability of diagnostic results. These shortcomings collectively restrict the practical application of power battery leakage detection technology, necessitating the development of a novel leakage detection method capable of real-time online monitoring, early warning, low false alarm rates, and multi-parameter collaborative analysis. Summary of the Invention

[0005] The technical problem solved by this invention is to provide a power battery leakage detection method and system that can realize real-time online detection, integrate multi-parameter collaborative analysis, have early warning capabilities, and effectively reduce false alarm rate.

[0006] The basic solution provided by this invention is a method for detecting leakage in a power battery, comprising the following steps: S1. Real-time acquisition of battery cell voltage sequence and insulation resistance value in the battery management system, and cleaning of missing and outlier values ​​in the battery cell voltage sequence; S2. Standardize the voltage sequence of the cleaned battery elevator and extract the minimum voltage value and voltage standard deviation according to the preset time window to construct a two-dimensional feature vector; S3. Input the two-dimensional feature vector into a multivariate exponentially weighted moving average model to calculate the abnormal accumulation amount, and perform a moving average smoothing process on the abnormal accumulation amount. If the smoothed abnormal accumulation amount exceeds the first preset threshold, trigger a level one alarm. S4. After the Level 1 alarm is triggered, locate the lowest voltage point within the alarm period, and extract a local voltage window sequence centered on the store. Calculate the Z-score value of the lowest voltage point and the proportion of the lowest voltage value in the total data. The Z-score value is used to measure the degree of local deviation. S5. A leak is considered to have occurred when the following conditions are met simultaneously: The Z score exceeds a preset second preset threshold; The insulation resistance value is lower than a third preset threshold; The proportion of the lowest voltage occurrence exceeds the fourth preset threshold; S6. Generate a leakage risk score based on the accumulated abnormal amount and the Z score, and output the leakage alarm result and fault location information.

[0007] The principle and advantages of this invention are as follows: First, the voltage sequence and insulation resistance of individual battery cells in the battery management system are acquired in real time, and the voltage sequence is cleaned of missing and outlier values ​​to remove invalid and outlier points in the data to ensure the accuracy of subsequent analysis. Next, the cleaned voltage sequence is standardized to eliminate the influence of dimensions. Then, the minimum voltage value and standard deviation of voltage within each preset time window are extracted, and these two features are combined into a two-dimensional feature vector. This vector can simultaneously reflect the overall voltage fluctuation level and extreme low values. The constructed two-dimensional feature vector is input into a multivariate exponentially weighted moving average model. This model can smooth the multivariate time series data and calculate the cumulative amount of anomalies. The cumulative amount of anomalies is then smoothed using a moving average to eliminate random fluctuation interference. When the smoothed cumulative amount of anomalies exceeds a first preset threshold, a level one alarm is triggered, achieving preliminary screening of potential leakage risks. After the level one alarm is triggered, the lowest voltage point within the alarm period is further located, and the Z-score value of this point is calculated by extracting a local voltage window sequence. Simultaneously, the proportion of this lowest voltage value in the total data is statistically analyzed. The Z-score measures the degree of deviation of the voltage point within the local window, while the proportion reflects the prevalence of the outlier. A leak is determined to have occurred when all three conditions are met simultaneously: the Z-score exceeds the second preset threshold, the insulation resistance is below the third preset threshold, and the proportion of minimum voltage occurrences exceeds the fourth preset threshold. This multi-condition joint determination method can effectively eliminate false alarms from a single parameter. Finally, a leak risk score is generated based on the abnormal accumulation and the Z-score, and the leak alarm result and fault location information are output.

[0008] Existing technologies typically rely on a single parameter, such as a sudden drop in insulation resistance, or manual disassembly for detection. These methods suffer from issues such as detection lag, difficulty in capturing early, weak signals, and susceptibility to environmental interference leading to false alarms. This new method, through multivariate collaborative analysis combined with a hierarchical verification mechanism, can detect coordinated abnormal changes in voltage and insulation resistance in the early stages of leakage, enabling early warning. Furthermore, the multi-condition joint judgment significantly reduces the false alarm rate. The non-invasive online detection method avoids the destructive nature of traditional disassembly and detection, thus improving the overall real-time performance and reliability of power battery leakage detection.

[0009] Furthermore, S1 includes the following steps: S11. Real-time acquisition of data from the battery management system (BMS), including timestamps, battery elevator voltage sequences, and insulation resistance values; S12. Clean the collected data, delete missing values ​​and abnormal voltage data that exceed the preset reasonable range, and obtain the cleaned operating data.

[0010] Furthermore, S2 includes the following steps: S21. Divide the cleaned operation data into preset time windows, and calculate the minimum value and standard deviation of the battery cell voltage in each time window. S22. Standardize the voltage data within each time window to obtain standardized voltage data; S23. Combine the minimum voltage value and the standard deviation of voltage within each time window to form a two-dimensional feature vector. ,in This is the minimum voltage value. This represents the standard deviation of voltage.

[0011] First, the cleaned operational data is divided into several data blocks according to a preset time window. The size of the time window can be set according to actual needs, such as five minutes or ten minutes, dividing the continuous data stream into several data blocks for processing. For the battery cell voltage data within each time window, two key statistics are calculated: the minimum voltage and the standard deviation of all cell voltages within that window. The minimum voltage reflects the worst-performing cell voltage state within that time window, while the standard deviation reflects the consistency of voltage across all cells. Then, the voltage data within each time window is standardized. Standardization transforms the data into a distribution with a mean of zero and a standard deviation of 1 by subtracting the mean and dividing by the standard deviation, eliminating the influence of differences in voltage dimensions and amplitudes under different operating conditions on subsequent analysis. Finally, the calculated minimum voltage and standard deviation for each time window are combined to form a two-dimensional feature vector.

[0012] Furthermore, S3 includes the following steps: S31, Set the smoothness coefficient ,in And obtain the covariance matrix S based on historical data:

[0013] in Let i be the i-th historical two-dimensional feature vector. is the mean of the historical feature vectors, and n is the number of historical samples; S32, Two-dimensional feature vector for each time window Perform iterative calculations to obtain the MEWMA vector. :

[0014] Wherein, the initial vector t is the time window sequence; S33. Calculate the multivariate vector based on the MEWMA vector. Statistic As an abnormal accumulation:

[0015] in, Covariance matrix:

[0016] S34, regarding the aforementioned multi-element Statistic A moving average smoothing process is performed to obtain the smoothed cumulative anomaly. :

[0017] Where w is the size of the sliding window.

[0018] A smoothing coefficient, ranging from 0 to 1, is set to control the weighting ratio of historical and current data in the model. The covariance matrix is ​​estimated based on historical data, reflecting the correlation between the minimum voltage value and the standard deviation of voltage in the two-dimensional eigenvectors. Then, the two-dimensional eigenvectors for each time window are iteratively calculated to obtain the MEWMA vector. The initial vector is set to zero. The iterative formula weights the current eigenvector with the MEWMA vector from the previous time step, enabling the model to exponentially smooth the data. A multivariate statistic is calculated as the anomaly accumulation based on the calculated MEWMA vector. This statistic is derived from the quadratic form of the MEWMA vector and the inverse of its covariance matrix, comprehensively measuring the deviation of the current multivariate state from the normal state. Finally, the multivariate statistic sequence is smoothed using a moving average, taking the average of the statistics from the most recent several time windows to further eliminate random fluctuations and short-term disturbances, resulting in the smoothed anomaly accumulation. Existing univariate control charts cannot simultaneously monitor multiple related variables and lack sensitivity to weak anomaly signals. This step employs a multivariate exponentially weighted moving average method, which simultaneously monitors two relevant variables: the minimum voltage value and the standard deviation. It utilizes the covariance matrix to capture the correlation between variables, improving sensitivity to weak fault signals. The exponential weighting mechanism ensures that historical information continues to influence current judgments, while the moving average smoothing reduces noise interference, enabling the effective identification of weak abnormal signals caused by early leakage.

[0019] Furthermore, S4 includes the following steps: S41. Obtain the smoothed cumulative anomaly calculated in S3. ; S42, The smoothed abnormal accumulation amount With the first preset threshold If a comparison is made, This triggers a Level 1 alarm, indicating a potential risk of leakage and initiating a local anomaly depth verification process.

[0020] Furthermore, S5 includes the following steps: S51. During the period when the location triggers the Level 1 alarm, the minimum voltage of a single battery cell. and its corresponding time index ; S52, indexed by the time Centered on a preset window length, a sequence of local voltage windows is extracted. Where L is the half width of the window; S53. Calculate the minimum voltage value. Z-score in a voltage window sequence:

[0021] in, The mean of the local voltage window sequence, The standard deviation of the local voltage window sequence; S54. Calculate the minimum voltage value. Proportion of occurrence in total sampling points :

[0022] in, The voltage value is equal to The number of sampling points This represents the total number of sampling points.

[0023] First, locate the minimum voltage of a single battery cell within the period that triggers the Level 1 alarm and its corresponding time index; that is, find the cell with the lowest voltage value and the time of occurrence within the alarm period. Then, using this time index as the center, extract a local voltage window sequence with a preset window length. The window length is set according to actual needs, for example, taking several sampling points before and after, forming a local data segment centered on the lowest voltage point. Calculate the Z-score of this lowest voltage point in the local voltage window sequence. The Z-score is calculated by subtracting the mean of the local window from the lowest voltage value and then dividing by the standard deviation of the local window. This value reflects the degree of deviation of the lowest voltage point from the local data segment; the larger the Z-score, the more significant the abnormality of the point. Simultaneously, calculate the ratio of the frequency of the lowest voltage value in the total number of sampling points to the total number of sampling points, i.e., the occurrence ratio. This ratio reflects whether the abnormal voltage value is an occasional fluctuation or a persistent abnormal state. Current technologies often lack in-depth verification of abnormal points, judging anomalies solely based on voltage falling below a certain threshold, easily misinterpreting normal fluctuations as faults. This step, through local window Z-score calculation, can accurately measure the significance of outliers within a local range, eliminating interference from overall data fluctuations. The statistical analysis of occurrence rates can distinguish between occasional noise and persistent anomalies; if the proportion of a certain lowest voltage value is very low, it indicates a rare abnormal jump, more likely a genuine fault than random fluctuation. This local-depth verification method makes subsequent comprehensive judgments more accurate and reliable.

[0024] Furthermore, S5 also includes the following steps: S55, compare the Z score with the second preset threshold. The insulation resistance value is compared with a third preset threshold. Compare the proportions of the lowest voltage occurrences. With the fourth preset threshold Compare, when both Z > R < , If the leak is confirmed, it is considered a Level 1 alarm state; otherwise, it is considered a Level 1 alarm state.

[0025] Furthermore, S6 includes the following steps: S61, Abnormal accumulation after smoothing Combined with the Z-score, a leakage risk score is generated:

[0026] in, This represents the maximum value of the smoothed cumulative anomaly in the historical data. The maximum Z-score in the historical data. and These are the weighting coefficients, and they satisfy α+β=1; S62. If the first-level alarm is not triggered, output the normal status; If a Level 1 alarm is triggered but the leakage determination condition in step S55 is not met, a Level 1 alarm will be output, prompting that manual review is required. If the leakage determination condition of step S55 is met, a level 2 alarm is output to indicate that a leakage has occurred; S63. Perform visualization marking, draw voltage-time curve and insulation resistance decay curve, mark voltage jump points in the voltage-time curve, and mark insulation resistance decrease points in the insulation resistance decay curve.

[0027] First, a leakage risk score is generated by fusing the smoothed cumulative anomaly amount and the Z-score. The cumulative anomaly amount reflects the degree of long-term cumulative anomaly of multiple variables, while the Z-score reflects the significance of local voltage jumps. Both describe the fault state from the perspectives of macro-trend and local characteristics, respectively. A comprehensive risk quantification index is obtained by weighting and fusing the two. The weighting coefficients are set according to actual needs to adjust the contribution ratio of the two indicators in the final score. Then, corresponding alarm levels are output according to different judgment states. If a Level 1 alarm is not triggered, a normal state is output, indicating that the current battery condition is good. If a Level 1 alarm is triggered but the leakage judgment condition is not met, a Level 1 alarm is output, prompting manual review. At this point, the system has detected an abnormal signal but has not yet reached the standard for confirming a leakage. If the leakage judgment condition is met, a Level 2 alarm is output, indicating that a leakage has been confirmed and immediate action is required. Finally, visualization is performed by plotting voltage-time curves and insulation resistance decay curves. Voltage jump points are marked on the voltage-time curve, and insulation resistance decrease points are marked on the insulation resistance decay curve, visually displaying the time point of the fault and the abnormal characteristics in a graphical way.

[0028] This invention also discloses a power battery leakage detection system, and the aforementioned power battery leakage detection method includes: The data acquisition and preprocessing module is used to acquire the battery cell voltage sequence and insulation resistance value in the battery management system in real time, and to clean the missing and outlier values ​​of the battery cell voltage sequence. The feature extraction and construction module is used to standardize the voltage sequence of the cleaned battery cells and extract the minimum voltage value and voltage standard deviation according to a preset time window to construct a two-dimensional feature vector. The multivariate initial screening module is used to input the two-dimensional feature vector into the multivariate exponential weighted moving average model, calculate the cumulative abnormality, and perform a moving average smoothing process on the cumulative abnormality. If the smoothed cumulative abnormality exceeds a first preset threshold, a level one alarm is triggered. The local verification module is used to locate the lowest voltage point during the alarm period after the first-level alarm is triggered, and to extract a local voltage window sequence centered on the point, calculate the Z-score value of the lowest voltage point and the proportion of the lowest voltage value in the total data. The Z-score value is used to measure the degree of local deviation. The comprehensive judgment module is used to determine that leakage has occurred when the Z score exceeds the second preset threshold, the insulation resistance is lower than the third preset threshold, and the proportion of the lowest voltage occurrence exceeds the fourth preset threshold. The risk scoring and output module is used to generate a leakage risk score based on the accumulated abnormality and the Z score, and output leakage alarm results and fault location information. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of an embodiment of a power battery leakage detection method according to the present invention. Detailed Implementation

[0030] The following detailed description illustrates the specific implementation method: The basic implementation examples are as follows: Figure 1 As shown: A method for detecting leakage in a power battery includes the following steps: S1. Real-time acquisition of battery cell voltage sequence and insulation resistance value in the battery management system, and cleaning of missing and outlier values ​​in the battery cell voltage sequence.

[0031] S1 includes the following steps: S11. Real-time acquisition of data from the battery management system (BMS), including timestamps, battery elevator voltage sequences, and insulation resistance values; S12. Clean the collected data, delete missing values ​​and abnormal voltage data that exceed the preset reasonable range, and obtain the cleaned operating data.

[0032] Specifically, taking a ternary lithium-ion power battery pack as an example, this battery pack consists of 96 individual cells connected in series, with a rated voltage of 350V. During actual vehicle operation, BMS data is collected in real time via the CAN bus, with a sampling frequency set to 1 Hz, meaning one set of data is recorded per second. The collected data includes the timestamp corresponding to each data point, the voltage value of each of the 96 individual cells, and the insulation resistance between the positive terminal of the battery pack and the vehicle ground. The normal operating voltage range of an individual cell is 2.5V to 4.2V; below 2.5V is considered over-discharge, and above 4.2V is considered over-charge, both of which are abnormal data. During actual data collection, due to CAN bus communication interference, the voltage value of cell number 23 was recorded as 0V in one sampling, which is significantly lower than the lower limit of 2.5V and is considered an abnormal value, thus being deleted during the cleaning process. Simultaneously, during continuous operation, several data points had complete timestamps but missing values ​​among the 96 voltage values; for example, the voltage value of cell number 45 was missing in one data point, and this type of data was also deleted. After cleaning, a continuous and valid operational dataset was obtained, containing approximately 5000 valid data points. Each data point includes a complete timestamp, 96 voltage values, and the corresponding insulation resistance value. Table 1 below shows a comparison of some data before and after cleaning: Table 1

[0033] The above cleaning process eliminated invalid data caused by communication interference and sensor malfunctions, ensuring the quality of data used in subsequent analysis.

[0034] S2. Standardize the voltage sequence of the cleaned battery elevator and extract the minimum voltage value and voltage standard deviation according to the preset time window to construct a two-dimensional feature vector.

[0035] S2 includes the following steps: S21. Divide the cleaned operation data into preset time windows, and calculate the minimum value and standard deviation of the battery cell voltage in each time window. S22. Standardize the voltage data within each time window to obtain standardized voltage data; S23. Combine the minimum voltage value and the standard deviation of voltage within each time window to form a two-dimensional feature vector: ,in This is the minimum voltage value. This represents the standard deviation of voltage.

[0036] Specifically, the time window length is set to 300 seconds, or 5 minutes. This is because voltage anomalies caused by battery leakage typically don't occur within seconds but persist for a period. A 5-minute window captures the anomaly trend without obscuring local features due to an excessively large window. The cleaned, valid data is divided into several consecutive 300-second windows in chronological order, with no overlap between adjacent windows. For each time window, 96 individual cell voltage values ​​from all sampling points within that window are read, and the minimum and standard deviation of all voltage values ​​within that window are calculated. For example, under normal vehicle driving conditions, within a 5-minute window, the voltage values ​​of the 96 individual cells fluctuate between 3.75V and 3.95V. The calculated minimum voltage value for that window is 3.75V, and the standard deviation is 0.05V. Next, all voltage data within that window are standardized. The standardization formula is to subtract the mean of all voltage values ​​within the window from each voltage value, and then divide by the standard deviation. The standardized data has a mean of 0 and a standard deviation of 1. The purpose of standardization is to eliminate the impact of voltage amplitude differences under different operating conditions on subsequent analysis, making the data comparable under different driving conditions. Finally, the minimum voltage value and voltage standard deviation calculated by this window are combined into a two-dimensional feature vector, i.e. =[3.75,0.05] transpose. Table 2 below shows the feature extraction results for three consecutive time windows: Table 2

[0037] As can be seen from the table, the minimum voltage gradually decreases and the standard deviation of voltage gradually increases over time, indicating that the consistency of the battery pack deteriorates and there may be potential faults.

[0038] S3. Input the two-dimensional feature vector into the multivariate exponential weighted moving average model to calculate the abnormal accumulation amount, and perform a moving average smoothing process on the abnormal accumulation amount. If the smoothed abnormal accumulation amount exceeds the first preset threshold, trigger a level one alarm.

[0039] S3 includes the following steps: S31, Set the smoothness coefficient ,in And obtain the covariance matrix S based on historical data:

[0040] in Let i be the i-th historical two-dimensional feature vector. is the mean of the historical feature vectors, and n is the number of historical samples; S32, Two-dimensional feature vector for each time window Perform iterative calculations to obtain the MEWMA vector. :

[0041] Wherein, the initial vector t is the time window sequence; S33. Calculate the multivariate vector based on the MEWMA vector. Statistic As an abnormal accumulation:

[0042] in, Covariance matrix:

[0043] S34, regarding the aforementioned multi-element Statistic A moving average smoothing process is performed to obtain the smoothed cumulative anomaly. :

[0044] Where w is the size of the sliding window.

[0045] Specifically, historical data collected over 24 hours of continuous operation of the battery pack in a healthy state was first selected, totaling approximately 288 time windows. The covariance matrix S of these 288 two-dimensional feature vectors was then calculated. The calculated covariance matrix S is [0.0004, 0.0001], [0.0001, 0.0003]. This matrix reflects the correlation between the minimum voltage and the standard deviation of voltage. The diagonal elements represent the variances of the two variables, and the off-diagonal elements represent the covariance. A smoothing coefficient λ was set to 0.2. This coefficient determines the weight ratio of historical data to current data in the model. The smaller the λ, the stronger the model's dependence on historical data and the less sensitive it is to short-term fluctuations. Initial MEWMA vector. Set it as the zero vector. For each time window, according to... =0.2× +0.8× The MEWMA vector is calculated iteratively using the formula. For example, when the feature vector of the first window is [3.75, 0.05], =0.2×[3.75,0.05]+0.8×[0,0]=[0.75,0.01]. When the feature vector of the second window is [3.73,0.06], =0.2×[3.73,0.06]+0.8×[0.75,0.01]=[1.346,0.02], and so on. Then, the multivariate T² statistic is calculated based on the MEWMA vector, and this statistic is obtained through... transpose multiplied by The inverse of the covariance matrix multiplied by Received, among which The covariance matrix is ​​calculated from λ and S as (λ / (2-λ)). S. Set the moving average window size w to 3, that is, take the average of the T² statistic over three consecutive time windows to obtain the smoothed cumulative anomaly. By analyzing historical normal data, set a first preset threshold. The threshold is 12. A Level 1 alarm is triggered when the smoothed cumulative anomaly exceeds 12. During actual vehicle operation, at the 45th time window, the calculated smoothed cumulative anomaly was 13.5, exceeding the threshold of 12. The system then triggered a Level 1 alarm, indicating a potential leakage risk, and proceeded to the next verification stage. Table 3 below shows the changes in the cumulative anomaly from window 40 to window 46: Table 3

[0046] S4. After the first-level alarm is triggered, locate the lowest voltage point within the alarm period, and extract a local voltage window sequence centered on the store. Calculate the Z-score value of the lowest voltage point and the proportion of the lowest voltage value in the total data. The Z-score value is used to measure the degree of local deviation.

[0047] S4 includes the following steps: S41. Obtain the smoothed cumulative anomaly calculated in S3. ; S42, The smoothed abnormal accumulation amount With the first preset threshold If a comparison is made, This triggers a Level 1 alarm, indicating a potential risk of leakage and initiating a local anomaly depth verification process.

[0048] Specifically, after triggering a Level 1 alarm in the 45th window, the system immediately enters the local anomaly deep verification step. First, it locates the minimum voltage of 96 individual cells across all sampling points within the time period corresponding to the 45th window, along with their corresponding time indices. Data query reveals that within the time period corresponding to the 45th window (10:45:00 to 10:49:59), the voltage of cell number 78 at 10:47:23 is 3.52V, the lowest among all voltage values ​​within that time period. Centered on this time index 10:47:23, 60 sampling points are extracted before and after it, resulting in a local voltage window sequence of 121 sampling points spanning 60 seconds before and after 10:47:23. The mean and standard deviation of this local window sequence are calculated; the mean is 3.71V, and the standard deviation is 0.08V. Then, the Z-score for the lowest voltage point of 3.52V within this local window is calculated: Z = (3.52 - 3.71) / 0.08 = -2.375, and the absolute value is 2.375. Simultaneously, the frequency of the lowest voltage value of 3.52V in the total number of sampling points is counted. Of the 5000 valid data points, there are 5 sampling points with a voltage value equal to 3.52V, representing a frequency of 0.001 (one in a thousand). Table 4 below shows some data from the local window sequence: Table 4

[0049] S5. A leak is considered to have occurred when the following conditions are met simultaneously: The Z score exceeds a preset second preset threshold; The insulation resistance value is lower than a third preset threshold; The proportion of the lowest voltage occurrence exceeds the fourth preset threshold.

[0050] S5 includes the following steps: S51. During the period when the location triggers the Level 1 alarm, the minimum voltage of a single battery cell. and its corresponding time index ; S52, indexed by the time Centered on a preset window length, a sequence of local voltage windows is extracted. Where L is the half width of the window; S53. Calculate the minimum voltage value. Z-score in a voltage window sequence:

[0051] in, The mean of the local voltage window sequence, The standard deviation of the local voltage window sequence; S54. Calculate the minimum voltage value. Proportion of occurrence in total sampling points :

[0052] in, The voltage value is equal to The number of sampling points This represents the total number of sampling points.

[0053] S55, compare the Z score with the second preset threshold. The insulation resistance value is compared with a third preset threshold. Compare the proportions of the lowest voltage occurrences. With the fourth preset threshold Compare, when both Z > R < , If the leak is confirmed, it is considered a Level 1 alarm state; otherwise, it is considered a Level 1 alarm state.

[0054] Specifically, based on historical normal data and statistical experience, a second preset threshold is set. The value is 3, the third preset threshold. The insulation safety standard for battery systems is set at 100 ohms per volt. This battery pack has a rated voltage of 350V and corresponds to the following insulation resistance threshold. 350 multiplied by 100 equals 35000 ohms, or 35kΩ, the fourth preset threshold. The value is set to 0.0005, or five ten-thousandths. In the preceding steps, the calculated Z-score is 2.375, while... The value is 3, and 2.375 is less than 3, so the Z-score condition is not met. Querying the insulation resistance data for this period, during the 45th window, the insulation resistance gradually decreased from the normal 500kΩ to 380kΩ, still above the 35kΩ threshold, thus failing to meet the insulation resistance condition. (Lowest voltage occurrence ratio) It is 0.001, which is greater than The ratio of 0.0005 satisfies the occurrence ratio condition. Since two of the three conditions are not met, the system does not confirm a leak but maintains a Level 1 alarm status, prompting for manual verification. If, in subsequent testing, the verification data after a Level 1 alarm simultaneously meets the following conditions: Z-score greater than 3, insulation resistance less than 35kΩ, and minimum voltage occurrence ratio greater than 0.0005, then a leak is confirmed. Table 5 below shows the determination results for the three conditions: Table 5

[0055] S6. Generate a leakage risk score based on the accumulated abnormality and the Z score, and output the leakage alarm result and fault location information. S6 includes the following steps: S61, Abnormal accumulation after smoothing Combined with the Z-score, a leakage risk score is generated:

[0056] in, This represents the maximum value of the smoothed cumulative anomaly in the historical data. The maximum Z-score in the historical data. and These are the weighting coefficients, and they satisfy α+β=1; S62. If the first-level alarm is not triggered, output the normal status; If a Level 1 alarm is triggered but the leakage determination condition in step S55 is not met, a Level 1 alarm will be output, prompting that manual review is required. If the leakage determination condition of step S55 is met, a level 2 alarm is output to indicate that a leakage has occurred; S63. Perform visualization marking, draw voltage-time curve and insulation resistance decay curve, mark voltage jump points in the voltage-time curve, and mark insulation resistance decrease points in the insulation resistance decay curve.

[0057] Specifically, the weighting coefficients α and β are first set to 0.6 and β to 0.4, respectively, meaning the cumulative abnormality accounts for 60% of the risk score, and the Z-score accounts for 40%, with the sum of the two equaling 1. Based on historical data, the maximum smoothed cumulative abnormality is approximately 25, and the maximum Z-score is approximately 6. In the 45th window, the smoothed cumulative abnormality is 13.5, and the Z-score is 2.375. Therefore, the risk score is calculated as 0.6 multiplied by 13.5 divided by 25 plus 0.4 multiplied by 2.375 divided by 6, which equals 0.6 multiplied by 0.54 plus 0.4 multiplied by 0.396, which equals 0.324 plus 0.158, which equals 0.482. The maximum risk score is 0.482, out of 1. According to the alarm level classification, since the leakage judgment criteria were not met after this Level 1 alarm, the system outputs a Level 1 alarm, prompting manual review. Simultaneously, the system provides visual output, plotting the voltage-time curve from 10:00:00 to 10:50:00. A red dot marks the voltage jump point of 3.52V for cell number 78 at 10:47:23, and an insulation resistance decay curve is plotted, with blue dots marking the trend of the insulation resistance decreasing from 500kΩ to 380kΩ during this period. After receiving a Level 1 alarm, maintenance personnel can view the visual curve and, based on manual judgment, decide whether further disassembly and inspection are necessary. If subsequent windows meet the leakage detection criteria, the system will output a Level 2 alarm, confirming a leak and requiring immediate action. The table below shows the output content for different alarm levels:

[0058] This invention also discloses a power battery leakage detection system, and the aforementioned power battery leakage detection method includes: The data acquisition and preprocessing module is used to acquire the battery cell voltage sequence and insulation resistance value in the battery management system in real time, and to clean the missing and outlier values ​​of the battery cell voltage sequence. The feature extraction and construction module is used to standardize the voltage sequence of the cleaned battery cells and extract the minimum voltage value and voltage standard deviation according to a preset time window to construct a two-dimensional feature vector. The multivariate initial screening module is used to input the two-dimensional feature vector into the multivariate exponential weighted moving average model, calculate the cumulative abnormality, and perform a moving average smoothing process on the cumulative abnormality. If the smoothed cumulative abnormality exceeds a first preset threshold, a level one alarm is triggered. The local verification module is used to locate the lowest voltage point during the alarm period after the first-level alarm is triggered, and to extract a local voltage window sequence centered on the point, calculate the Z-score value of the lowest voltage point and the proportion of the lowest voltage value in the total data. The Z-score value is used to measure the degree of local deviation. The comprehensive judgment module is used to determine that leakage has occurred when the Z score exceeds the second preset threshold, the insulation resistance is lower than the third preset threshold, and the proportion of the lowest voltage occurrence exceeds the fourth preset threshold. The risk scoring and output module is used to generate a leakage risk score based on the accumulated abnormality and the Z score, and output leakage alarm results and fault location information.

[0059] The above are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A method for detecting leakage in a power battery, characterized in that: Includes the following steps: S1. Real-time acquisition of battery cell voltage sequence and insulation resistance value in the battery management system, and cleaning of missing and outlier values ​​in the battery cell voltage sequence; S2. Standardize the voltage sequence of the cleaned battery elevator and extract the minimum voltage value and voltage standard deviation according to the preset time window to construct a two-dimensional feature vector; S3. Input the two-dimensional feature vector into a multivariate exponentially weighted moving average model to calculate the abnormal accumulation amount, and perform a moving average smoothing process on the abnormal accumulation amount. If the smoothed abnormal accumulation amount exceeds the first preset threshold, trigger a level one alarm. S4. After the Level 1 alarm is triggered, locate the lowest voltage point within the alarm period, and extract a local voltage window sequence centered on the store. Calculate the Z-score value of the lowest voltage point and the proportion of the lowest voltage value in the total data. The Z-score value is used to measure the degree of local deviation. S5. A leak is considered to have occurred when the following conditions are met simultaneously: The Z score exceeds a preset second preset threshold; The insulation resistance value is lower than a third preset threshold; The proportion of the lowest voltage occurrence exceeds the fourth preset threshold; S6. Generate a leakage risk score based on the accumulated abnormal amount and the Z score, and output the leakage alarm result and fault location information.

2. The method for detecting leakage in a power battery according to claim 1, characterized in that: S1 includes the following steps: S11. Real-time acquisition of data from the battery management system (BMS), including timestamps, battery elevator voltage sequences, and insulation resistance values; S12. Clean the collected data, delete missing values ​​and abnormal voltage data that exceed the preset reasonable range, and obtain the cleaned operating data.

3. The method for detecting leakage in a power battery according to claim 2, characterized in that: S2 includes the following steps: S21. Divide the cleaning operation data into preset time windows, and calculate the minimum value and standard deviation of the battery cell voltage in each time window. S22. Standardize the voltage data within each time window to obtain standardized voltage data; S23. Combine the minimum voltage value and the standard deviation of voltage within each time window to form a two-dimensional feature vector. ,in This is the minimum voltage value. This represents the standard deviation of voltage.

4. The method for detecting leakage in a power battery according to claim 3, characterized in that: S3 includes the following steps: S31, Set the smoothness coefficient ,in And obtain the covariance matrix S based on historical data: in Let i be the i-th historical two-dimensional feature vector. is the mean of the historical feature vectors, and n is the number of historical samples; S32, Two-dimensional feature vector for each time window Perform iterative calculations to obtain the MEWMA vector. : Wherein, the initial vector t is the time window sequence; S33. Calculate the multivariate vector based on the MEWMA vector. Statistic As an abnormal accumulation: in, Covariance matrix: S34, regarding the aforementioned multi-element Statistic Perform a moving average smoothing process to obtain the smoothed cumulative anomaly. : Where w is the size of the sliding window.

5. The method for detecting leakage in a power battery according to claim 4, characterized in that: S4 includes the following steps: S41. Obtain the smoothed cumulative anomaly calculated in S3. ; S42, The smoothed abnormal accumulation amount With the first preset threshold If a comparison is made, This triggers a Level 1 alarm, indicating a potential risk of leakage and initiating a local anomaly depth verification process.

6. The method for detecting leakage in a power battery according to claim 5, characterized in that: S5 includes the following steps: S51. During the period when the location triggers the Level 1 alarm, the minimum voltage of a single battery cell. and its corresponding time index ; S52, indexed by the time Centered on a preset window length, a sequence of local voltage windows is extracted. Where L is the half width of the window; S53. Calculate the minimum voltage value. Z-score in a voltage window sequence: in, The mean of the local voltage window sequence, The standard deviation of the local voltage window sequence; S54. Calculate the minimum voltage value. Proportion of occurrence in total sampling points : in, The voltage value is equal to The number of sampling points This represents the total number of sampling points.

7. The method for detecting leakage in a power battery according to claim 6, characterized in that: S5 further includes the following steps: S55, compare the Z score with the second preset threshold. The insulation resistance value is compared with a third preset threshold. Compare the proportions of the lowest voltage occurrences. With the fourth preset threshold Compare, when both Z > R < , If the leak is confirmed, it is considered a Level 1 alarm state; otherwise, it is considered a Level 1 alarm state.

8. The method for detecting leakage in a power battery according to claim 7, characterized in that: S6 includes the following steps: S61, Abnormal accumulation after smoothing Combined with the Z-score, a leakage risk score is generated: in, This represents the maximum value of the smoothed cumulative anomaly in the historical data. The maximum Z-score in the historical data. and These are the weighting coefficients, and they satisfy α+β=1; S62. If the first-level alarm is not triggered, output the normal status; If a Level 1 alarm is triggered but the leakage determination condition in step S55 is not met, a Level 1 alarm will be output, prompting that manual review is required. If the leakage determination condition of step S55 is met, a level 2 alarm is output to indicate that a leakage has occurred; S63. Perform visualization marking, draw voltage-time curve and insulation resistance decay curve, mark voltage jump points in the voltage-time curve, and mark insulation resistance decrease points in the insulation resistance decay curve.

9. A power battery leakage detection system, employing the power battery leakage detection method according to any one of claims 1-8, characterized in that, include: The data acquisition and preprocessing module is used to acquire the battery cell voltage sequence and insulation resistance value in the battery management system in real time, and to clean the missing and outlier values ​​of the battery cell voltage sequence. The feature extraction and construction module is used to standardize the voltage sequence of the cleaned battery cells and extract the minimum voltage value and voltage standard deviation according to a preset time window to construct a two-dimensional feature vector. The multivariate initial screening module is used to input the two-dimensional feature vector into the multivariate exponential weighted moving average model, calculate the cumulative abnormality, and perform a moving average smoothing process on the cumulative abnormality. If the smoothed cumulative abnormality exceeds a first preset threshold, a level one alarm is triggered. The local verification module is used to locate the lowest voltage point during the alarm period after the first-level alarm is triggered, and to extract a local voltage window sequence centered on the point, calculate the Z-score value of the lowest voltage point and the proportion of the lowest voltage value in the total data. The Z-score value is used to measure the degree of local deviation. The comprehensive judgment module is used to determine that leakage has occurred when the Z score exceeds the second preset threshold, the insulation resistance is lower than the third preset threshold, and the proportion of the lowest voltage occurrence exceeds the fourth preset threshold. The risk scoring and output module is used to generate a leakage risk score based on the accumulated abnormality and the Z score, and output leakage alarm results and fault location information.

Images