New energy automobile power battery multi-fault diagnosis method

Abstract

The invention relates to the technical field of lithium ion battery fault diagnosis, and discloses a new energy automobile power battery multi-fault diagnosis method. According to the method, a double-layer diagnosis strategy is adopted, in the first layer, multi-layer feature extraction is conducted on battery voltage data, multiple abnormal conditions of abnormal single batteries in a battery pack are accurately recognized through an ALOF algorithm, and in the second layer, inner short circuit fault diagnosis is further conducted on the abnormal single batteries through the Mahalanobis distance. In order to extract fault features with finer granularity, kurtosis and change rates are extracted from different time windows of voltage data after the influence of environment and sensor errors is eliminated as indexes, and then preliminary anomaly screening is carried out on voltage changes of different time scales of each single battery by using an ALOF algorithm. Normal battery average voltage data which can reflect the current battery working condition and aging condition and is screened out after first-layer diagnosis is adopted as a comparison object, and the mahalanobis distance is used for further diagnosing the internal short circuit fault of the single battery which is diagnosed to be abnormal in the first layer.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium-ion battery fault diagnosis, and specifically to a multi-fault diagnosis method for power batteries of new energy vehicles. Background Art

[0002] The increasing ownership of new energy vehicles has led to serious safety problems of power batteries. The safety of power batteries is related to people's lives and property. Different from the spontaneous combustion of traditional fuel vehicles, due to the activity of lithium elements in power batteries, once thermal runaway occurs, the safety escape window is only a few minutes. At present, there is no effective method to solve this problem. Therefore, it is necessary to study how to locate and diagnose potential faults in the early stage of fault occurrence.

[0003] There are mainly three types of fault diagnosis methods in existing research: knowledge-based methods, model-based methods, and data-driven methods. Knowledge-based methods rely heavily on expert knowledge and have poor applicability. Model-based methods are difficult to achieve accurate diagnosis under real working conditions because the modeling uses experimental data. Data-driven methods use mainly real vehicle data, so they have more practical application value. In recent years, with the rise of big data platforms, the method combining platform vehicle operation data and data-driven has good real-time performance and strong historical data utilization ability. However, such methods still have the following problems at present: (1) The current diagnosis methods are difficult to efficiently diagnose multiple faults and anomalies. (2) Commonly used algorithms are difficult to accurately judge multiple anomalies and multiple adjacent anomalies, making the algorithms prone to missed diagnosis; (3) Existing algorithms are basically based on fixed data or thresholds for comparative diagnosis, without considering the serious influence of battery working conditions and aging degree, and are prone to missed diagnosis. Summary of the Invention

[0004] (I) Technical Problems to be Solved

[0005] In view of the deficiencies of the prior art, the present invention provides a multi-fault diagnosis method for power batteries of new energy vehicles, which has the advantages of identifying and locating multiple anomalies and multiple adjacent anomaly batteries, and solves the above technical problems.

[0006] (II) Technical Solutions

[0007] To achieve the above object, the present invention provides the following technical solution: A multi-fault diagnosis method for power batteries of new energy vehicles, comprising the following steps:

[0008] S1: Obtain the voltage data of all single cells in the real vehicle battery pack in the period before the sampling moment during operation, and construct a voltage matrix of the battery pack , represents the th single cell in the battery pack, represents Sub-sampling, taking as the voltage sequence at the detection moment, and performing preliminary anomaly detection on the voltage data at the detection moment according to the nominal upper and lower limit voltage values of the battery at the time of factory

[0009] S2: Using the voltage matrix including the voltage data of the previous 60 moments before the current moment to calculate the kurtosis of the voltage data of each single battery within the time window, and then combining the kurtosis values of all single batteries to generate a kurtosis feature vector ;

[0010] S3: Using the voltage data of the previous 6 moments before the current moment included in the voltage matrix to calculate the voltage change rate of the voltage data of each single battery within the time window, and then combining the voltage change rates of all single batteries to generate a voltage change rate feature vector ;

[0011] S4: Bringing the kurtosis feature vector , the voltage change rate feature vector and the voltage sequence at the detection moment into the ALOF algorithm for calculation, respectively obtaining the corresponding ALOF values, that is , , , judging , , whether any one of them exceeds the corresponding threshold , , , if it exceeds the corresponding threshold, execute step S5;

[0012] S5: Calculating the Mahalanobis distance M between the historical voltage sequence of the abnormal single battery and the average voltage of the normal batteries in the same period. When the single battery is greater than the threshold , it is determined that this single battery has an internal short circuit fault, and the specific expression is as follows:

[0013]

[0014] Among them, represents the Mahalanobis distance calculated between the historical voltage sequence of the abnormal single battery and the average voltage of the normal batteries in the same period, represents the historical voltage data of the abnormal single battery, and the superscript represents the transpose, represents the vector of the average historical voltage of the normal batteries, represents the historical voltage covariance matrix of the normal batteries, and its expression is as follows:

[0015]

[0016]

[0017] Among them, represents the voltage value at the b-th sampling point, represents the total number of voltage data.

[0018] As a preferred technical solution of the present invention, before constructing the voltage matrix in step S1, the acquired data is also preprocessed. The preprocessing algorithm uses the wavelet threshold denoising algorithm to perform wavelet decomposition on the original voltage data to obtain wavelet coefficients of each layer. The wavelet coefficient formula is:

[0019]

[0020] Among them, z represents the number of sampling points, represents taking the logarithm of the number of sampling points;

[0021] Subsequently, abnormal data points are judged and removed according to the threshold, and the soft threshold function is selected as the threshold function, and its expression is:

[0022]

[0023] Among them, is the wavelet coefficient, is the sign function, is the critical threshold.

[0024] As a preferred technical solution of the present invention, the kurtosis feature vector in the -th single cell kurtosis has the following specific expression:

[0025]

[0026]

[0027]

[0028] Among them, represents the kurtosis of the -th single cell, represents the window size selected for the long time scale, represents the voltage value of the -th sampling starting from the current moment in the time window, represents the average voltage within the selected window, represents the average of the voltage data of the -th single cell within the time window, represents the The standard deviation of the voltage data of each single cell in the time window.

[0029] As a preferred technical solution of the present invention, the voltage change rate characteristic vector Middle Voltage change rate of a single cell The calculation formula is:

[0030]

[0031] in, represents the window size selected for the short time scale, Indicates the voltage value at the left end of the time window, Indicates the voltage value at the current moment.

[0032] As a preferred technical solution of the present invention, the specific steps of the ALOF algorithm in step S4 are as follows:

[0033] S4.1: Calculate the reachable distance between objects y and x :

[0034] in, and They are objects and No. eigenvalues, is the total number of eigenvalues, Express The data are summed;

[0035] S4.2: Set an initial domain size ,object Within the area size The reachable distance is:

[0036]

[0037] At this time object The initial neighborhood It is expressed as:

[0038]

[0039] in, represents the entire data set, Representation Object Within the area size The reachable distance, Represents the distance between any object in domain k and object x;

[0040] S4.3: Calculation Objects The local density of

[0041]

[0042] where represents the average distance between object x and the objects in its initial neighborhood, which is used to reflect the local density, represents the number of objects contained in the initial neighborhood of object 𝑥, represents the sum of the distances between object 𝑥 and all objects y in its initial neighborhood.

[0043] S4.4: Calculate the average value of the local distances of all objects in the entire dataset of :

[0044]

[0045] refers to the total number of objects in the dataset.

[0046] Determine the dynamically changing neighborhood size corresponding to each object according to the local density and average local density of the data is:

[0047]

[0048] where is the proportionality coefficient, which is used to control the change range of the neighborhood size and is usually set to 3; the symbol represents rounding up to ensure is an integer; reflects the ratio of the local density of object 𝑥 to the global average density. If is small (i.e., the neighborhood density of 𝑥 is high), then will be large, and vice versa.

[0049] S4.5: For the new neighborhood value , the reachable distance of object x within the neighborhood size is:

[0050]

[0051] The new neighborhood value including object x itself within the neighborhood is represented as:

[0052]

[0053] The new local density is:

[0054]

[0055] Among them represents the number of objects contained in the corresponding dynamic neighborhood of object 𝑥, represents the reachable distance in the dynamically adjusted neighborhood:

[0056]

[0057] For object , repeating the same process with object can obtain ;

[0058] S4.6 The updated ALOF value after modification is:

[0059]

[0060] Among them, represents 's ALOF value;

[0061] The threshold , , 's specific expressions are as follows

[0062]

[0063] Among them, represents the eigenvalue , , one of them, is 's corresponding ALOF value, is the eigenvalue The mean value of all single cells value under, is all single cells under this eigenvalue The standard deviation of the value.

[0064] Compared with the prior art, the present invention provides a multi-fault diagnosis method for power batteries of new energy vehicles, having the following beneficial effects:

[0065] 1. By adopting voltage characteristics on different time scales, the present invention can capture battery abnormal phenomena more comprehensively and with finer granularity, and uses a double-layer diagnosis strategy combined with the ALOF algorithm and Mahalanobis distance to achieve accurate diagnosis of multiple faults.

[0066] 2. The present invention uses the improved LOF algorithm ALOF in the method of locating abnormal single cells. The algorithm can better identify and distinguish multiple adjacent abnormal objects by dynamically adjusting the domain size and k value, and can efficiently and accurately identify and locate abnormal single cells in the battery pack.

[0067] 3. By using the fault diagnosis method of the present invention, the average normal battery voltage data diagnosed by the first layer of the diagnosis strategy represents the general voltage state of normal batteries under this working condition, that is, the aging degree. Calculate the Mahalanobis distance between it and the historical voltage data of abnormal batteries, and then use a threshold to judge the fault situation of abnormal single batteries. The algorithm takes into account the influence of battery aging and working conditions on performance, which helps to reduce missed diagnoses. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a schematic flow chart of the present invention;

[0069] Figure 2 It is a schematic diagram of voltage-time data of a normal vehicle of the present invention;

[0070] Figure 3 It is a schematic diagram of voltage-time data of a faulty vehicle of the present invention;

[0071] Figure 4 It is a schematic diagram of voltage-time data of a normal vehicle after noise reduction of the present invention;

[0072] Figure 5 It is a schematic diagram of voltage-time data of a faulty vehicle after noise reduction of the present invention;

[0073] Figure 6 It is a schematic diagram of the ALOF result of a faulty vehicle of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0074] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0075] Please refer to Figures 1-6 , a multi-fault diagnosis method for power batteries of new energy vehicles, including the following steps:

[0076] S1: Obtain the voltage data of all single batteries in the real vehicle battery pack in the front section of the sampling moment during operation, and construct a voltage matrix of the battery pack , represents the th battery cell in the battery pack, represents times of sampling, and As the voltage sequence at the detection moment, and the voltage data at the detection moment is preliminarily detected for anomalies according to the nominal upper and lower limit voltage values of the battery at the time of factory shipment. The nominal voltage upper limit of the single battery used in the vehicle battery pack is 4.15V, and the nominal voltage lower limit is 2.75V;

[0077] Before constructing the voltage matrix in step S1, the acquired data is also preprocessed. The preprocessing algorithm uses the wavelet threshold denoising algorithm to perform wavelet decomposition on the original voltage data to obtain the wavelet coefficients of each layer. The wavelet coefficient formula is:

[0078]

[0079] where z represents the number of sampling points, represents taking the logarithm of the number of sampling points;

[0080] Subsequently, the abnormal data points are judged according to the threshold and the abnormal data is removed. The soft threshold function is selected as the threshold function, and its expression is:

[0081]

[0082] where, is the wavelet coefficient, is the sign function, is the critical threshold.

[0083] S2: Using the voltage matrix which includes the voltage data of the previous 60 moments before the current moment to calculate the kurtosis of the voltage data of each single battery within the time window, and then combining the kurtosis values of all single batteries to generate a kurtosis feature vector , the kurtosis feature vector The kurtosis of the th single battery in

[0084]

[0085]

[0086]

[0087] where, represents the kurtosis of the th single battery, represents the window size selected for the long time scale, represents the voltage value of the th sampling starting from the current moment in the time window, represents the average voltage within the selected window, represents the The mean value of the voltage data of the th single cell within the time window;

[0088] S3: Calculate the voltage change rate of the voltage data of each single cell within the time window by using the voltage data of the 6 moments before the current moment in the voltage matrix, and then combine the voltage change rates of all single cells to generate a voltage change rate feature vector , and the voltage change rate of the th single cell in the voltage change rate feature vector is calculated as follows:

[0089]

[0090] where, represents the window size selected for the short time scale, represents the voltage value at the leftmost end of the time window, represents the voltage value at the current moment;

[0091] S4: Substitute the kurtosis feature vector , the voltage change rate feature vector and the voltage sequence at the detection moment into the ALOF algorithm for calculation, and respectively obtain the corresponding ALOF values, that is , , , and judge whether any one of , , exceeds the corresponding threshold , , . If it exceeds the corresponding threshold, execute step S5. The specific steps of the ALOF algorithm are as follows:

[0092] S4.1: Calculate the reachable distance between object y and x :

[0093] where, and are respectively the th and th eigenvalue of objects , is the total number of eigenvalues, represents the sum of data;

[0094] S4.2: Set an initial neighborhood size , object​ The reachable distance within the domain size is:

[0095]

[0096] At this time, the initial neighborhood of the object is expressed as:

[0097]

[0098] Among them, represents the entire dataset, represents the object The reachable distance within the domain size is represents the distance between any object in the k-th domain and the object x;

[0099] S4.3: Calculate the local density of the object as:

[0100]

[0101] Among them, represents the average distance between the object x and the objects in its initial neighborhood, which is used to reflect the local density, represents the number of objects contained in the initial neighborhood of the object 𝑥, represents the sum of the distances between the object 𝑥 and all objects y in its initial neighborhood.

[0102] S4.4: Calculate the average value of the local distances of all objects in the entire dataset :

[0103]

[0104] refers to the total number of objects in the dataset.

[0105] Determine the dynamically changing neighborhood size corresponding to each object according to the local density and average local density of the data as:

[0106]

[0107] Among them, is the proportionality coefficient, which is used to control the change range of the neighborhood size and is usually set to 3; the symbol represents rounding up to ensure is an integer; reflects the ratio of the local density of the object 𝑥 to the global average density. If ​​​If it is smaller (i.e., the neighborhood density of 𝑥 is higher), then will be larger, and vice versa.

[0108] S4.5: For the new domain value , the reachable distance of the object x within the domain size is:

[0109]

[0110] The neighborhood including the object x itself under the new neighborhood value is represented as:

[0111]

[0112] The new local density is:

[0113]

[0114] where represents the number of objects contained in the corresponding dynamic neighborhood of the object 𝑥, represents the reachable distance in the dynamically adjusted neighborhood:

[0115]

[0116] For the object , repeating the same process with the object can obtain ;

[0117] The updated ALOF value after S4.6 modification is:

[0118]

[0119] where, represents 's ALOF value;

[0120] The threshold , , 's specific expressions are as follows:

[0121]

[0122] where, represents one of the eigenvalue , , , is 's corresponding ALOF value, is the eigenvalue for all single cells under The mean value of the values, for all single cells at this eigenvalue is the standard deviation of the values;

[0123] S5: Calculate the Mahalanobis distance M between the historical voltage sequence of the abnormal single cell and the mean voltage of the normal cells in the same period. When the is greater than the threshold then it is determined that this single cell has an internal short circuit fault, The value of is 1.12, and the specific expression is as follows:

[0124]

[0125] where, represents the Mahalanobis distance calculated between the historical voltage sequence of the th abnormal single cell and the mean voltage of the normal cells in the same period, represents the historical voltage data of the abnormal single cell, and the superscript represents the transpose, represents the mean voltage vector of the historical voltages of the normal cells, represents the historical voltage covariance matrix of the normal cells, and its expression is as follows:

[0126]

[0127]

[0128] where, represents the voltage value at the bth sampling point, represents the total number of voltage data;

[0129] Kurtosis is used in statistics to describe the shape of the data distribution and can reflect the steepness and tail thickness of the data distribution. Kurtosis is defined as the fourth-order central moment of a random variable divided by the fourth power of the standard deviation and is a dimensionless parameter. In battery fault diagnosis, the kurtosis value of the voltage data of each single cell in the 10 minutes before the diagnosis moment is calculated to characterize the comprehensive change degree of the battery voltage within a long time window;

[0130] The ALOF algorithm used in S4 is an algorithm optimized for the detection effect of multiple single cells having anomalies simultaneously and multiple adjacent single cells having anomalies simultaneously on the basis of the LOF algorithm. Its basic algorithm, LOF, is an unsupervised learning algorithm commonly used for anomaly detection. This algorithm identifies outliers by comparing the local density differences between the target data points and their adjacent data points. This algorithm is widely used in battery fault diagnosis. However, this algorithm is greatly affected by the selection of the neighborhood size k, and the diagnostic results are easily interfered by data noise. Most importantly, when multiple anomalies occur simultaneously and multiple anomalies occur adjacent to each other, the neighborhoods of this algorithm highly overlap, and it is easy to misjudge as normal. Therefore, in this solution, on the basis of the original algorithm, it is modified to dynamically adjust the neighborhood size k of each object according to the local density of the area where the object is located and include the object x itself in its neighborhood;

[0131] Using the ALOF algorithm to perform anomaly judgment on different voltage characteristic values can not only detect potential anomalies more accurately and comprehensively, but also diagnose some potential faults simultaneously; when the value exceeds the corresponding threshold, it indicates that there is an abnormal fluctuation in the voltage of the single cell; when the value exceeds the corresponding threshold, it indicates that the voltage of the single cell has an abnormal rise and fall recently; When the value exceeds the corresponding threshold, it indicates that there is a risk of inconsistency in the single cell;

[0132] In the further diagnosis of the located abnormal single cells using the Mahalanobis distance, all single cells in S4 where , , do not exceed the corresponding thresholds are used as the normal single cells during this period. The historical voltage data of them are averaged at each sampling moment to obtain the voltage sequence data that can represent the general state of normal batteries during this period. Using this as the comparison object for abnormal single cells can diagnose faults more accurately. Before step S5, the abnormal single cells have been located. This step further diagnoses the faults of the abnormal single cells. The Mahalanobis distance can accurately and efficiently calculate the difference between two sequences. Calculating the Mahalanobis distance between the historical voltage data of the abnormal single cell and the average voltage of the normal batteries can effectively diagnose the actual fault situation of the abnormal battery;

[0133] Specifically, in this embodiment, voltage-time data during the charging process of normal vehicles and faulty vehicles are obtained from the new energy vehicle big data platform. The voltage-time data of normal vehicles are as shown in Figure 2 , and the voltage-time data of faulty vehicles are as shown in Figure 3 ; Figure 4 shows the voltage-time data of the normal vehicle after noise reduction, Figure 5 shows the voltage-time data of the faulty vehicle after noise reduction.

[0134] For a faulty vehicle, select the time corresponding to the 3500th data point as the diagnosis time, and then construct a voltage matrix using historical data ; First, based on the upper and lower limits of the nominal voltage, preliminarily judge whether there is an abnormality or serious fault in the voltage sensor for the voltage data at the diagnosis time; then calculate the characteristics corresponding to each single battery respectively and , and obtain the corresponding feature vectors 、 , and then calculate 、 and the voltage sequence at the diagnosis time corresponding 、 、 . By comparing with the corresponding thresholds, the specific ALOF values are as shown in Figure 6 . It can be found that there is a risk of inconsistency in the 12th single battery, and there are abnormal voltage fluctuations in the 53rd, 54th, and 55th single batteries, and there have been abnormal voltage rises and falls recently; therefore, the 12th, 53rd, 54th, and 55th single batteries are determined to be abnormal batteries and further internal short-circuit diagnosis is required.

[0135] Select all single batteries except the 12th, 53rd, 54th, and 55th single batteries as the representatives of normal batteries during this period, calculate the voltage mean sequence of normal single batteries, and then calculate the Mahalanobis distances between the voltage mean sequence and the first 60 voltage data of these four single batteries, and the values are 1.71, 2.37, 2.11, and 2.25 respectively, exceeding the threshold of 1.12. Therefore, it is diagnosed that these four single batteries have all suffered internal short-circuit faults, and the deviation degrees of their voltage data from the voltage mean sequence of normal batteries are different. It can be understood that the single battery with a greater deviation from the voltage mean sequence of normal batteries is more likely to have a more serious fault degree, which is consistent with the observable deviation degree in the fault voltage curve graph.

[0136] The fault diagnosis method of the present invention is applicable to the real-time monitoring of battery status and multi-fault diagnosis under real working conditions. This method adopts a two-layer diagnosis strategy. In the first layer, multi-level feature extraction is performed on battery voltage data, and an improved Advanced Local Outlier Factor (ALOF) algorithm is used to accurately identify various abnormal conditions of abnormal single cells in the battery pack. In the second layer, the Mahalanobis Distance is used to further diagnose the internal short-circuit fault of the abnormal single cell battery. In order to extract fault features with finer granularity, kurtosis and change rate are respectively extracted as indicators for different time windows of the voltage data after eliminating the influence of environmental and sensor errors, and then the ALOF algorithm is used to preliminarily screen for abnormalities in the voltage changes of each single cell battery at different time scales, while diagnosing various abnormalities, locating the abnormal single cell battery; since the change of the battery with working conditions and aging will cause the change of the voltage change trend, fixed thresholds and comparison objects will lead to missed diagnoses, so the scheme uses the average voltage data of the normal batteries selected after the first layer of diagnosis that can reflect the current battery working conditions and aging conditions as the comparison object, and uses the Mahalanobis Distance to further diagnose the internal short-circuit fault of the abnormal single cell battery diagnosed in the first layer. The present invention takes into account the influence of the change of the battery with working conditions and aging conditions on fault diagnosis, has stronger adaptability in actual working condition applications, and at the same time, the two-layer diagnosis strategy integrating the improved outlier algorithm can realize multi-fault diagnosis and abnormal battery location, and the algorithm is efficient while effectively solving the problems of misdiagnosis and missed diagnosis of adjacent abnormal single cell batteries caused by the change of working conditions and aging conditions;

[0137] In summary, the new energy vehicle power battery fault diagnosis method provided by the embodiments of the present invention uses multi-time scale features to characterize the state of the battery. Combining with the improved ALOF algorithm, it can more accurately locate abnormal single cell batteries and distinguish various abnormal conditions; the battery fault diagnosis method further diagnoses the internal short-circuit fault of the abnormal single cell battery on the basis of abnormal single cell battery location. The Mahalanobis Distance algorithm based on the average voltage mean data of normal batteries in the same period can perform fault diagnosis more efficiently and accurately, reducing misdiagnosis. The overall scheme has the advantages of diagnosis efficiency, accuracy, and robustness.

[0138] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-fault diagnosis method for power batteries of new energy vehicles, characterized by: The following steps are involved: S1: Obtain the voltage data of all single cells in the real vehicle battery pack during operation before the sampling time, and construct the voltage matrix of the battery pack , Represents the battery pack Single battery, represent Sub-sampling, As the voltage sequence at the detection time, the voltage data at the detection time is subjected to preliminary abnormality detection according to the nominal upper and lower voltage limits of the battery when it leaves the factory; S2: Using voltage matrix The voltage data of the 60 moments before the current moment are included in the calculation of the kurtosis of the voltage data of each single cell in the time window, and then the kurtosis values ​​of all single cells are combined to generate the kurtosis feature vector ; S3: Use the voltage data of the six moments before the current moment in the voltage matrix to calculate the voltage change rate of each single cell in the time window, and then combine the voltage change rates of all single cells to generate a voltage change rate feature vector ; S4: The kurtosis eigenvector , voltage change rate characteristic vector and the voltage sequence at the detection time Bring it into the ALOF algorithm for calculation and get the corresponding ALOF values, namely , , ,judge , , Whether any of the items exceeds the corresponding threshold , , , if it exceeds the corresponding threshold, execute step S5; S5: Calculate the Mahalanobis distance M between the historical voltage sequence of the abnormal single cell and the mean voltage of the normal cells in the same period. Greater than threshold When , it is determined that this single battery has an internal short circuit fault. The specific expression is as follows: in, The Mahalanobis distance between the historical voltage sequence of abnormal single cells and the mean voltage of normal cells in the same period is calculated. Indicates the historical voltage data of abnormal single cells, with superscript represents transpose, represents the historical voltage mean vector of a normal battery, Represents the historical voltage covariance matrix of a normal battery, and its expression is as follows: in, Indicates the voltage value of the bth sampling point, Indicates the total number of voltage data.

2. The multi-fault diagnosis method for power batteries of new energy vehicles according to claim 1 is characterized in that: In step S1, before constructing the voltage matrix, the acquired data is preprocessed. The preprocessing algorithm uses a wavelet threshold noise reduction algorithm to perform wavelet decomposition on the original voltage data to obtain the wavelet coefficients of each layer, wherein the wavelet coefficient formula is: Where z represents the number of sampling points, It means taking the logarithm of the number of sampling points; Then, the abnormal data points are judged according to the threshold and the abnormal data is removed. The soft threshold function is selected As a threshold function, its expression is: in, is the wavelet coefficient, is the symbolic function, is the critical threshold.

3. The multi-fault diagnosis method for power batteries of new energy vehicles according to claim 1 is characterized in that: The kurtosis eigenvector Middle The kurtosis of a single cell The specific expression is as follows: in, Indicates The kurtosis of a single cell, represents the window size selected for a long time scale, Indicates the time window starting from the current time. The voltage value of the subsample, represents the mean voltage in the selected window, Indicates The average value of the voltage data of each single cell in the time window, Indicates The standard deviation of the voltage data of each single cell in the time window.

4. The multi-fault diagnosis method for power batteries of new energy vehicles according to claim 1 is characterized in that: The voltage change rate characteristic vector Middle Voltage change rate of a single cell The calculation formula is: in, represents the window size selected for the short time scale, Indicates the voltage value at the left end of the time window, Indicates the voltage value at the current moment.

5. The multi-fault diagnosis method for power batteries of new energy vehicles according to claim 1 is characterized in that: The specific steps of the ALOF algorithm in step S4 are as follows: S4.1: Calculate the reachable distance between objects y and x : in, and They are objects and No. eigenvalues, is the total number of eigenvalues, Express The data are summed; S4.2: Set an initial domain size ,object Within the area size The reachable distance is: At this time object The initial neighborhood It is expressed as: in, represents the entire data set, Representation Object Within the area size The reachable distance, Represents the distance between any object in domain k and object x; S4.3: Calculation Objects The local density is: in, Represents the average distance between object x and the objects in its initial neighborhood, which is used to reflect the local density. represents the number of objects contained in the initial neighborhood of object 𝑥, represents the sum of the distances between object 𝑥 and all objects y in its initial neighborhood; S4.4: Compute local distances for all objects in the entire dataset The average : in, refers to the total number of objects in the dataset, It means sum; Determine the dynamically changing neighborhood size corresponding to each object based on the local density and average local density of the data for: in, is the proportional coefficient, which is used to control the change of the neighborhood size. Indicates rounding up to ensure is an integer; S4.5: For new field values , object x within the domain size The reachable distance is: New neighborhood value The neighborhood including the object x itself is represented as: New local density for: in represents the number of objects contained in the corresponding dynamic neighborhood of object 𝑥, represents the reachable distance in the dynamically adjusted neighborhood, Indicates that all To sum, Represents the new neighborhood value The neighborhood of the object x itself is expressed as follows: For objects , repetition and object The same process can be used to obtain the new local density ; S4.6 The modified ALOF value is updated as follows: in, express ALOF value; The threshold , , The specific expression is as follows: in, Representative eigenvalue , , One of them, for The corresponding ALOF value is is the eigenvalue All single cells The mean of the values, For all single cells under this characteristic value The standard deviation of the values.

Images