A new energy automobile battery voltage fault diagnosis method based on operation data

Abstract

The application relates to the field of electric vehicle battery fault diagnosis methods, in particular to a new energy vehicle battery voltage fault diagnosis method based on operation data, which carries out cleaning treatment on battery signal data to obtain time, charging and discharging data and a voltage matrix; time and the voltage matrix meeting preset conditions are reserved; feature extraction is carried out on the voltage matrix, and a feature matrix D is obtained by calculating each row of data; the feature matrix D is subjected to sliding mean filter processing to obtain a voltage feature extraction matrix; the voltage feature extraction matrix is subjected to sliding calculation, the features of each cell in the window are summed and calculated, and then differential calculation is carried out to obtain a feature value trend change matrix; according to time points, the feature vectors of the voltage feature extraction matrix and the feature value trend change matrix are traversed, whether the feature values meet the conditions is judged, if yes, it is judged as suspected abnormality; if suspected abnormality occurs at the same time in the same cell, it is determined that the final abnormality occurs. The application reduces the false positive rate of fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle battery fault diagnosis methods, specifically to a new energy vehicle battery voltage fault diagnosis method based on operational data. Background Technology

[0002] Abnormal battery voltage signals are highly correlated with faults such as battery self-discharge and abnormal battery system connections. Severe voltage signal anomalies can ultimately lead to battery thermal runaway, posing a significant threat to user property and safety. Currently, online voltage warnings generally rely on threshold settings. These methods have relatively broad thresholds, poor detection timeliness, and are susceptible to the influence of data sensor sampling failures, battery connection abnormalities, and other related faults on the voltage signal, greatly increasing the false alarm rate for continuously abnormal voltage drops. Furthermore, traditional battery fault detection primarily relies on offline equipment for diagnosis, depending on prior experience and rules. This results in poor adaptability to unexpected faults, limitations imposed by detection equipment, and poor application to large-scale data. Data-driven prediction methods can accurately predict battery faults by starting with historical battery data and utilizing statistical principles and intelligent technologies. However, existing prediction methods are prone to problems such as untimely voltage anomaly warnings and numerous false alarms. Summary of the Invention

[0003] The present invention aims to provide a method for diagnosing battery voltage faults in new energy vehicles based on operational data, so as to solve the problems of untimely early warning of abnormal voltage faults and a large number of false alarms.

[0004] The battery voltage fault diagnosis method for new energy vehicles based on operational data in this solution includes:

[0005] Step S1: Parse the original message to extract battery signal data;

[0006] Also includes:

[0007] Step S2: Clean the battery signal data to obtain the time, charge / discharge state, and voltage matrix;

[0008] Step S3: Calculate the voltage data of the charging state. The preset condition for calculation is CHARGE_STATUS == 1. The time and voltage matrices corresponding to the preset condition are retained.

[0009] Step S4: Sample and extract features from the data in the voltage matrix, and use the median voltage difference method to calculate the feature matrix D for each row of data.

[0010] Step S5: Perform moving mean filtering on the feature matrix D based on a specific window to obtain the voltage feature extraction matrix K;

[0011] Step S6: Select a specific window to perform sliding calculation on the voltage feature extraction matrix K, sum the features of each cell within the window and then perform difference calculation to obtain the feature value trend change matrix J;

[0012] Step S7: Based on the time point, traverse the feature vector of each cell in the voltage feature extraction matrix K and the feature value trend change matrix J, and determine whether its feature value is less than the lower limit of the abnormal threshold or greater than the upper limit of the abnormal threshold. If the condition is met, it is judged as a suspected abnormality.

[0013] Step S8: Simultaneously judge the voltage feature extraction matrix K and the feature value trend change matrix J. If a suspected anomaly occurs at the same time in the same cell, it is determined that the anomaly has finally occurred.

[0014] The beneficial effects of this plan are:

[0015] By extracting a feature matrix from the battery signal data after cleaning, and then calculating and extracting voltage characteristics and feature value trend changes, upper and lower thresholds are calculated for both voltage characteristics and feature value trend changes to preliminarily identify suspected anomalies. Finally, by combining voltage characteristics and feature value trend changes, it is determined whether the suspected anomaly occurred at the same time in the same battery cell; if so, the anomaly is confirmed. Based on features and corresponding trend changes, voltage anomaly fault warnings are more timely, reducing the false alarm rate.

[0016] Furthermore, in step S2, the cleaning process includes deleting abnormal characters and invalid data, removing data outside a preset range, and selecting data of a preset type.

[0017] Furthermore, in step S3, the preset condition is CHARGE_STATUS == 1.

[0018] Furthermore, in step S4, the voltage matrix is ​​represented as follows: , each column This represents the voltage data of the i-th cell, each row. Let V represent the voltage data at time point j. With a sampling interval of T seconds, the time corresponding to time point j is jT seconds. Feature extraction is performed on the voltage matrix V, and median voltage difference processing is performed on the voltage data V to obtain the feature matrix D of median voltage difference.

[0019] Furthermore, in step S4, for each row Calculate the median Then use each column Subtract the median This yields a new row vector as the feature matrix D, represented as:

[0020] ,in, .

[0021] The beneficial effect is that calculating the feature matrix D using the median makes the data more uniform.

[0022] Furthermore, in step S5, the moving average filtering process involves using a sliding window of length M, moving one time point at a time, and for each window k, calculating the average value of each column within it to obtain a new row vector, represented as:

[0023] ;

[0024] in, , where m is the row number of the characteristic matrix D.

[0025] The beneficial effect is that using moving average filtering can make the data smoother.

[0026] Further, in step S6, a sliding window calculation is performed on the voltage feature extraction matrix K. The length of the sliding window is 3*M. The sliding window moves by one time point each time, and there are a total of m-3*M+1 windows. For each window k, the sum of each column inside it is calculated to obtain a new row vector as follows:

[0027] ;

[0028] in, , This represents the sum of the nth column;

[0029] Then, perform difference operations on each column, that is, replace the original value with the difference between two adjacent rows, and finally obtain a new matrix representing the trend change of the eigenvalues. , is represented as:

[0030] ;

[0031] in, , .

[0032] The beneficial effect is that by setting a larger sliding window for calculation, the corresponding characteristic information in the voltage can be extracted and retained.

[0033] Further, in step S7, data from the first tenth of the entire life cycle are selected from the voltage feature extraction matrix K and the eigenvalue trend change matrix J. The entire life cycle includes L time points, and the first tenth of the life cycle is the first L / 10 time points, denoted as K_1 and J_1, both of which are (L / 10) rows and N columns. The column vectors of K_1 and J_1 are traversed, and the 75th quantile and 25th quantile of each column feature vector are obtained, denoted as Q3 and Q1 respectively. They are two 1-row n-column vectors.

[0034] The beneficial effect is that different quantiles are applied to the data of the first tenth of the entire life cycle of matrices K and J, ensuring the validity and completeness of the battery characteristics contained in the calculated data.

[0035] Furthermore, in step S7, for the i-th column, it is first sorted by time into an ascending sequence S_i, and then Q3 and Q1 are calculated according to the following formula:

[0036] ;

[0037] ;

[0038] in, .

[0039] The beneficial effects are: by performing formula calculations for each individual battery cell, the fault characterization information of each cell can be taken into account, avoiding the situation where insufficient abnormal data leads to warning failure; at the same time, this feature extraction method is based on the idea of ​​anomaly detection in statistics, which can quickly and accurately extract key information features of the battery cell.

[0040] Furthermore, in step S7, the upper threshold of the voltage feature extraction matrix K is calculated according to the quartile method. and lower threshold For the i-th column, the upper and lower thresholds are calculated according to the following formulas:

[0041] ;

[0042] ;

[0043] Each point in the voltage feature extraction matrix K is evaluated, and if it exceeds the upper limit threshold... and lower threshold The abnormal points are marked as 1, and the remaining normal points are marked as 0, resulting in a matrix K1 with values ​​ranging from 0 to 1.

[0044] The upper threshold of the eigenvalue trend change matrix J is calculated using the quartile method. and lower threshold Specifically, for the i-th column, it is calculated according to the following formula:

[0045] ;

[0046] .

[0047] The beneficial effect is that it judges each point of the voltage feature extraction matrix J, and if it exceeds the upper limit threshold... and lower threshold The abnormal points are marked as 1, and the remaining normal points are marked as 0, resulting in a matrix J1 with values ​​ranging from 0 to 1. Attached Figure Description

[0048] Figure 1 This is a flowchart of an embodiment of the new energy vehicle battery voltage fault diagnosis method based on operational data of the present invention;

[0049] Figure 2 This is a feature value map of each cell extracted during online verification in an embodiment of the new energy vehicle battery voltage fault diagnosis method based on operational data of the present invention;

[0050] Figure 3 This is a diagram showing the online verification and early warning results in an embodiment of the new energy vehicle battery voltage fault diagnosis method based on operational data of the present invention.

[0051] Figure 4 This is a graph showing the voltage results of each individual cell during online verification in an embodiment of the new energy vehicle battery voltage fault diagnosis method based on operational data according to the present invention.

[0052] Figure 5 This is the voltage diagram of cell number 105 during online verification of an embodiment of the new energy vehicle battery voltage fault diagnosis method based on operational data according to the present invention. Detailed Implementation

[0053] The following detailed description provides further details on specific implementation methods.

[0054] Example

[0055] New energy vehicle battery voltage fault diagnosis methods based on operational data, such as Figure 1 As shown, it includes:

[0056] Step S1 involves parsing the original message data as specified in the GB32960 national standard to extract the battery signal data. The original message data refers to the raw data (binary data) uploaded by the vehicle to the remote platform via the onboard communication module. Parsing this data into a format conforming to the GB32960 standard is a complex process that requires parsing according to the protocol specified in the standard.

[0057] The following are the general steps for parsing raw message data into GB32960 data format:

[0058] 1. GB32960 Standard: Based on the GB32960 standard, the message format, data items, and verification algorithms defined in the standard are determined. Documentation for the GB32960 standard can be obtained from the relevant standardization organization or official website.

[0059] 2. Parsing the raw message header: The GB32960 standard specifies the format of the message header, including fields such as the start character, command identifier, response identifier, and data unit length. When parsing the raw message, the header information must be parsed according to the standard definition.

[0060] 3. Parsing Data Units: Data units are the portions of the GB32960 message that contain actual vehicle data. The format and content of data units vary depending on the command identifier. Each data item, including vehicle speed, vehicle status, location information, and battery status, needs to be parsed according to the data unit format defined in the standard.

[0061] 4. Checksum and Verification: The GB32960 standard requires that a checksum be included in the message to ensure the integrity and accuracy of the data. During the parsing process, the checksum needs to be calculated and compared with the checksum in the message to verify the correctness of the data.

[0062] 5. Generate GB32960 format data: After parsing, the parsed data items are reassembled into data conforming to the GB32960 standard.

[0063] The actual parsing process may vary depending on factors such as vehicle model and equipment type. Generally, manufacturers will provide corresponding parsing tools or documentation to help developers implement the parsing process from raw messages to GB32960 data format, for example, to obtain the results shown in Table 1.

[0064] Table 1 Analysis Results

[0065]

[0066] Step S2: Clean the battery signal data to obtain time, charge / discharge data and voltage matrix. The cleaning process includes deleting abnormal characters and invalid data, removing data outside the preset range, and selecting data of preset types. Abnormal characters and invalid data include NAN, etc. The preset range is voltage data that is greater than 6V or less than 1V.

[0067] Step S3: Calculate the voltage data of the discharge state, and retain the time and voltage matrix corresponding to the conditions that are met. The preset condition is Charge_status==1. That is, the voltage data that meets the conditions is filtered according to the preset condition Charge_status==1.

[0068] Step S4: Sample and extract features from the data in the voltage matrix. Calculate the feature matrix D for each row of data using the median voltage difference method. The voltage matrix is ​​represented as follows: , each column This represents the voltage data of the i-th cell, each row. Let V represent the voltage data at time point j, with a sampling interval of T seconds. The time corresponding to time point j is jT seconds. Feature extraction is performed on the voltage matrix V, and median voltage difference processing is applied to the voltage data V to obtain the feature matrix D of the median voltage difference. For each row... Calculate the median Then use each column Subtract the median This yields a new row vector as the feature matrix D, represented as:

[0069] ,in, .

[0070] Step S5: Perform moving average filtering on the feature matrix D based on a specific window to obtain the voltage feature extraction matrix K. Then, plot the feature values ​​for each cell as shown below. Figure 2 The results shown have the x-axis representing time points, where the interval between two time points represents the time interval after the time data is extracted in step 2 (step size), and the y-axis representing feature values. The moving average filtering process involves using a sliding window of length M, moving one time point at a time. For each window k, the average value of each column within it is calculated, resulting in a new row vector, represented as:

[0071] ;

[0072] in, m is the row number of the feature matrix D. The sliding window can only start from the first row and end at the (m-M+1)th row, otherwise it will exceed the range of the matrix. When k=m-M+1, it means that the right endpoint of the sliding window just reaches the last row of the feature matrix. In other words, this is the last valid sliding window.

[0073] Step S6: Select a specific window to perform sliding calculation on the voltage feature extraction matrix K. Summate the features of each cell within the window and then perform difference calculation to obtain the feature value trend change matrix J. Perform sliding window calculation on the voltage feature extraction matrix K. The length of the sliding window is 3*M. The sliding window moves by one time point each time, and there are a total of M-3*M+1 windows. For each window k, calculate the sum of each column within it to obtain a new row vector:

[0074] ;

[0075] in, , This represents the sum of the nth column;

[0076] Then, perform difference operations on each column, that is, replace the original value with the difference between two adjacent rows, and finally obtain a new matrix representing the trend change of the eigenvalues. , is represented as:

[0077] ;

[0078] in, , .

[0079] Step S7: Based on the time point, traverse the feature vector of each cell in the voltage feature extraction matrix K and the feature value trend change matrix J. The time point is the time data obtained after parsing the original data in step 1. Determine whether its feature value is less than the lower limit of the abnormal threshold or greater than the upper limit of the abnormal threshold. If the condition is met, it is judged as a suspected abnormality.

[0080] We select the voltage feature extraction matrix K and the eigenvalue trend change matrix J as the data for the first tenth of the entire life cycle. The entire life cycle includes L time points, and the first tenth of the life cycle is the first L / 10 time points, denoted as K_1 and J_1, both of which are (L / 10) rows and N columns. We traverse the column vectors of K_1 and J_1 and find the 75th quantile and 25th quantile of each column feature vector, denoted as Q3 and Q1, respectively. They are two 1-row n-column vectors.

[0081] For the i-th column, first sort it into an ascending sequence S_i, then calculate Q3 and Q1 according to the following formula:

[0082] ;

[0083] ;

[0084] in, n is the column number, which refers to the number of battery cells. Here, Q3 and Q1 need to be calculated separately for each battery cell.

[0085] The upper threshold of the voltage feature extraction matrix K is calculated using the quartile method. and lower threshold For the i-th column, the upper and lower thresholds are calculated according to the following formulas:

[0086] ;

[0087] ;

[0088] The upper threshold of the eigenvalue trend change matrix J is calculated using the quartile method. and lower threshold Specifically, for the i-th column, it is calculated according to the following formula:

[0089] ;

[0090] .

[0091] Step S8: Simultaneously judge the voltage feature extraction matrix K and the eigenvalue trend change matrix J. If a suspected anomaly occurs at the same time in the same cell, it is determined that an anomaly has finally occurred. The cell is then marked as potentially experiencing a voltage anomaly at this time. Figure 3 As shown, the horizontal axis represents time, the vertical axis represents the extracted features, the long vertical line represents the warning position, and the remaining curves in the figure represent the changing trends of each cell's features over time. Specifically, the fault is mainly determined by multiplying two 0 / 1 matrices, K1 and J1. If both K1 and J1 are determined to be faulty, then the final fault matrix L is also determined to be faulty, resulting in a more accurate judgment.

[0092] In step S8, the fault anomalies are judged simultaneously. If the same cell (i.e., the same column) in two matrices is judged as abnormal at the same time (i.e., the same row), then it is ultimately determined to be abnormal. The anomaly judgment matrix is ​​as follows:

[0093] .

[0094] The upper and lower thresholds are calculated for the voltage feature extraction matrix K and the eigenvalue trend change matrix J, respectively. The thresholds are set separately for different battery cells, which is more targeted and takes into account the trend changes. This avoids false alarms caused by data fluctuations due to factors such as abnormal data sampling and control strategies. Ultimately, it enables timely and accurate identification of battery anomalies during traversal, improving the accuracy of diagnostic detection.

[0095] To verify the effectiveness of the above methods, online data verification analysis and offline data verification were conducted.

[0096] (1) Online verification

[0097] Taking the self-discharge anomaly of cell number 105 on 2021-01-20 03:29:17 as an example, the above method was used to analyze and confirm that the vehicle has a risk of battery self-discharge. A characteristic diagram of the self-discharge anomaly fault vehicle is shown below. Figure 3 As shown, the result of obtaining the feature matrix K is as follows: Figure 2 As shown, the voltage diagram of a single battery cell is as follows: Figure 4 As shown, the voltage diagram of cell number 105 is as follows. Figure 5 As shown in the figures, cell number 105 in the battery pack exhibited a significant outlier in its characteristic values, consistent with the characteristic curve of a voltage fault. Combined with the significant voltage drop, a preliminary assessment indicated self-discharge. After offline verification and repair measures, the characteristic values ​​returned to normal.

[0098] (2) Offline data verification

[0099] Battery cell number 105 in the battery pack experienced self-discharge. Offline disassembly and testing revealed significant abnormal capacity degradation in cell number 105, noticeable module deformation, and an abnormal drop in online voltage data, confirming that cell number 105 was experiencing a self-discharge fault. The offline testing results were consistent with the data analysis and algorithm model results.

[0100] There is a high correlation between abnormal battery voltage signals and battery self-discharge and abnormal battery system connections. This is mainly because battery voltage signals are one of the important parameters reflecting battery status, and these faults usually affect the normal operation and performance of the battery.

[0101] 1. Abnormal Battery Voltage Signal: Battery voltage is a crucial indicator of battery status, reflecting its charging state and energy reserves. Abnormal battery voltage signals can be caused by internal battery malfunctions, overcharging, over-discharging, or other issues. When the battery voltage signal is abnormal, the system may be unable to accurately determine the battery's actual state, leading to erroneous battery management control.

[0102] 2. Battery Self-Discharge: Battery self-discharge refers to the phenomenon where a battery gradually loses its stored charge and discharges current when there is no external load. Battery self-discharge may be caused by material problems within the battery itself, internal chemical reactions, etc. Abnormal battery voltage signals may cause the system to mistakenly believe that the battery is in a high-discharge state, thereby increasing the risk of battery self-discharge.

[0103] 3. Battery System Connection Anomalies: Battery system connection anomalies may include communication problems between the battery and the battery management system (BMS), or faulty connection lines. When the battery system connection is abnormal, the BMS may be unable to properly acquire various battery parameters, including the battery voltage signal. This can lead to the system's inability to accurately monitor and manage the battery status, increasing the risk of battery instability.

[0104] Therefore, abnormal battery voltage signals are closely related to battery self-discharge and abnormal battery system connections. On the one hand, these abnormalities may lead to abnormal battery voltage signals; on the other hand, abnormal battery voltage signals may also be one of the causes of battery self-discharge and connection abnormalities. Therefore, in a battery management system, timely monitoring and handling of abnormal battery voltage signals is an important measure to ensure the normal operation of the battery system and extend battery life.

[0105] This embodiment extracts a feature matrix from the battery signal data after cleaning, then calculates and extracts voltage features and feature value trend changes. Upper and lower thresholds are calculated for both voltage features and feature value trend changes to preliminarily identify suspected anomalies. Finally, by combining voltage features and feature value trend changes, it is determined whether the suspected anomaly occurred at the same time in the same battery cell; if so, the anomaly is confirmed. Based on features and corresponding trend changes, voltage anomaly fault warnings are more timely, reducing the false alarm rate.

[0106] The above descriptions are merely embodiments of the present invention, and common knowledge regarding specific structures and characteristics is not elaborated upon here. It should be noted that those skilled in the art can make various modifications and improvements without departing from the structure of the present invention, and these should also be considered within the scope of protection of the present invention. These modifications and improvements will not affect the effectiveness 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 diagnosing battery voltage faults in new energy vehicles based on operational data, comprising: Step S1: Parse the original message to extract battery signal data; Its characteristic is that it further includes: Step S2: Clean the battery signal data to obtain time, charge / discharge data, and voltage matrix; Step S3: Calculate the voltage data of the discharge state. The preset condition for calculation is Charge_status==1. The time and voltage matrix corresponding to the preset condition are retained. Step S4: Sample and extract features from the data in the voltage matrix, and use the median voltage difference method to calculate the feature matrix D for each row of data. The voltage matrix is ​​represented as , each column This represents the voltage data of the i-th cell, each row. This represents the voltage data at time point j, with a sampling interval of T seconds. The time corresponding to time point j is jT seconds. Feature extraction is performed on the voltage matrix V, and median voltage difference processing is applied to the voltage data V to obtain the feature matrix D of the median voltage difference. For each row... Calculate the median Then use each column Subtract the median This yields a new row vector as the feature matrix D, represented as: ,in, ; Step S5: Perform moving average filtering on the feature matrix D based on a specific window to obtain the voltage feature extraction matrix K. The moving average filtering process is as follows: using a moving window of length M, moving one time point at a time, for each window k, calculate the average value of each column inside it to obtain a new row vector, represented as: ； in, ; Step S6: Select a specific window to perform sliding calculation on the voltage feature extraction matrix K. Summate the features of each cell within the window and then perform difference calculation to obtain the feature value trend change matrix J. Perform sliding window calculation on the voltage feature extraction matrix K. The length of the sliding window is 3*M. The sliding window moves by one time point each time, and there are a total of m-3*M+1 windows. For each window k, calculate the sum of each column within it to obtain a new row vector: ； in, , This represents the sum of the i-th column; Then, perform difference operations on each column, that is, replace the original value with the difference between two adjacent rows, and finally obtain a new matrix representing the trend change of the eigenvalues. , represented as: ； in, , ; Step S7: Based on the time point, traverse the feature vector of each cell in the voltage feature extraction matrix K and the feature value trend change matrix J, and determine whether its feature value is less than the lower limit of the abnormal threshold or greater than the upper limit of the abnormal threshold. If the condition is met, it is judged as a suspected abnormality. Step S8: Simultaneously judge the voltage feature extraction matrix K and the feature value trend change matrix J. If a suspected abnormality occurs at the same time in the same cell, it is determined that the abnormality has finally occurred.

2. The method for diagnosing battery voltage faults in new energy vehicles based on operational data according to claim 1, characterized in that: In step S2, the cleaning process includes deleting abnormal characters and invalid data, removing data outside a preset range, and selecting data of a preset type.

3. The method for diagnosing battery voltage faults in new energy vehicles based on operational data according to claim 1, characterized in that: In step S3, the preset condition is Charge_status==1.

4. The method for diagnosing battery voltage faults in new energy vehicles based on operational data according to claim 1, characterized in that: In step S7, the voltage feature extraction matrix K and the eigenvalue trend change matrix J are selected based on the data of the first tenth stage of the entire life cycle. The entire life cycle includes L time points, and the first tenth stage is the first L / 10 time points, denoted as K_1 and J_1, both of which are (L / 10) rows and N columns. The column vectors of K_1 and J_1 are traversed, and the 75th quantile and 25th quantile of each column feature vector are obtained, denoted as Q3 and Q1 respectively. They are two 1-row n-column vectors.

5. The method for diagnosing battery voltage faults in new energy vehicles based on operational data according to claim 4, characterized in that: In step S7, for the i-th column, it is first sorted into an ascending sequence. Then calculate Q3 and Q1 according to the following formula: ； ； in, .

6. The method for diagnosing battery voltage faults in new energy vehicles based on operational data according to claim 5, characterized in that: In step S7, the upper threshold of the voltage feature extraction matrix K is calculated using the quartile method. and lower threshold For the i-th column, the upper and lower thresholds are calculated according to the following formulas: ； ； The upper threshold of the eigenvalue trend change matrix J is calculated using the quartile method. and lower threshold Specifically, for the i-th column, it is calculated according to the following formula: ； 。

Images