A rapid sorting method for the consistency of power batteries

Abstract

The present invention discloses a method for quickly sorting the consistency of power batteries. The power batteries are detected by the AC impedance test method, and the EIS test data of the batteries is dimensionally reduced to extract characteristic quantities for reflecting the battery performance, which helps to accelerate the sorting speed. Regarding the sorting problem of batteries, an adaptive clustering sorting method is proposed. This method uses the density of data to obtain the number of clusters and the initial values of the cluster centers to achieve preliminary sorting of the batteries, and then optimizes the sorting results according to the distances from the data points to the cluster centers to achieve further sorting. Existing methods often require a large number of battery samples for pre-training or need to determine the number of battery clusters in advance for sorting, with low test efficiency and inapplicable to the sorting of large-scale batteries. When sorting the batteries, the present invention has a relatively fast sorting speed, the batteries in the same group have good consistency, and the proposed method has good applicability to different types of batteries.

Description

Technical Field

[0001] The present invention belongs to the technical fields of electric vehicles and energy storage, and particularly relates to a method for quickly sorting the consistency of power batteries. Background Art

[0002] In 2020, the cumulative total of retired power batteries in China reached approximately 2 million tons, and this figure will rise to approximately 7.8 million tons by 2025. Against this background, the retired batteries of electric vehicles can provide considerable economic benefits through secondary utilization such as energy storage. However, the screening and recombination of large-scale power battery monomers face problems of low efficiency and low accuracy.

[0003] There are mainly four existing methods for detecting and sorting the consistency of power batteries: (1) By detecting the appearance (such as bulging and leakage), weight, size, and sealing of scrapped batteries to determine whether the batteries can be reused. (2) Using the test method of constant current charge and discharge to detect the capacity, internal resistance, etc. of power batteries one by one. However, such methods are time-consuming and not suitable for large-scale battery sorting. (3) Sorting based on battery test curves, such as using EIS, incremental capacity (IC) curves, pulse curves, etc. as the basis for battery sorting. However, there is a large amount of data redundancy in the battery test curves, which will reduce the sorting speed of the data. (4) Machine learning methods, including support vector machines and artificial neural networks (ANN), etc., can model multi-variable problems of complex systems and extract implicit non-linear relationships between variables. However, neural networks require a large amount of training data, so they are not suitable for battery sorting with fewer samples. In battery sorting, the number of classifications is a difficult point. If the number of classifications is too large, the sorting speed will be too slow; if the number of classifications is too small, the consistency of the sorted battery monomers will be poor.

[0004] The existing battery sorting methods have a slow test speed and are not suitable for sorting large-scale batteries. This method is based on the EIS test data of batteries, and by extracting the characteristic quantities related to battery performance in the EIS curve as the basis for battery sorting, it helps to accelerate the sorting speed; during clustering sorting, the density of data points is used for pre-sorting to preliminarily determine the number of clusters and the centers of the batteries, and by adjusting the distance between the data points and the center points, the optimization of the battery sorting results is realized, thereby realizing the adaptive selection of the number of classifications of different types of batteries, and there is no need to pre-train the test data of the battery samples to be sorted. While accelerating the sorting speed, it also improves the applicability of the sorting method to various types of batteries and ensures the consistency of the batteries within the sorted groups. Summary of the Invention

[0005] To solve the above technical problems, the present invention provides a method for rapid sorting of the consistency of power batteries. Due to differences in battery production processes or the usage environments after assembly, etc., there are differences in performance indicators such as battery internal resistance, capacity, open-circuit voltage, and Coulomb efficiency. Using the power battery sorting method described above, battery groups with similar performance can be screened out, ensuring the maximization of the battery life and the safety of use when the battery is applied again.

[0006] The present invention adopts the following technical solutions to solve the technical problems:

[0007] A method for rapid sorting of the consistency of power batteries includes the following steps:

[0008] S1. For the battery samples to be tested, use the alternating current impedance method for testing, obtain the electrochemical impedance spectrum data set of the batteries in the battery samples to be tested, and obtain the battery electrochemical impedance spectrum curve;

[0009] S2. Perform data dimensionality reduction on the electrochemical impedance spectrum data set of the battery samples to be tested, and use the data after dimensionality reduction to represent the batteries to be tested;

[0010] S3. Based on the data after dimensionality reduction, pre-sort the batteries. According to the adaptive clustering algorithm, use the data density as an index to obtain the number of clusters and the cluster centers, and as the iteration progresses, the density is adaptively updated;

[0011] S4. Further sort the batteries, and adjust the cluster centers by calculating the distances between the data points and the cluster centers.

[0012] Furthermore, the specific process of obtaining the battery electrochemical impedance spectrum curve in S1 is as follows:

[0013] Step (1). Apply excitation current signals with different frequencies to the battery. The excitation current value is taken as 1 / 20 of the rated capacity value of the battery, and synchronously sample the voltage signal U(t) and current signal I(t) of the battery;

[0014] Step (2). Through Fourier decomposition, extract the amplitudes of the voltage signal U(t) and current signal I(t) of the battery at different frequencies;

[0015] (14)

[0016] In the formula, ω is the angular frequency corresponding to the frequency f, are the phases corresponding to the voltage and current; n represents the nth injected frequency, and t represents the time interval;

[0017] Step (3). According to the voltage signal U(t) and current signal I(t) of the battery, calculate the amplitude Z(n) and phase of the battery impedance under the excitation signal :

[0018] (15)

[0019] Step (4). Change the frequency of the excitation signal, and repeat steps (2) and (3) to fit the electrochemical impedance spectrum curve of the battery from the impedance of the battery at different frequencies.

[0020] Furthermore, in S2, the data dimensionality reduction is to sort and weight the eigenvalues based on eigenvalue decomposition to reduce the number of feature quantities. The specific process is as follows:

[0021] The number of battery samples is m, and each battery is tested at n frequencies. The corresponding electrochemical impedance spectrum curve contains n points. Therefore, the original battery data is expressed as:

[0022] (16)

[0023] In the formula, X represents the set of all original battery data, and x m represents the column vector formed by the test data of the m-th battery.

[0024] Construct the covariance matrix B of X n,m :

[0025] (17)

[0026] Perform eigenvalue decomposition on the covariance matrix B to obtain the eigenvalues λ i , and the corresponding eigenvectors v i ; Arrange the eigenvectors v i in descending order by column from left to right according to the corresponding eigenvalue magnitudes, and take the first q columns to form the matrix T q，m ;

[0027] The contribution degree R i of the eigenvalue is defined as:

[0028] (18)

[0029] In the formula, λ i is the eigenvalue of the matrix T q，m ;

[0030] In order to determine the selected feature quantities and prevent the influence of high-dimensional data on subsequent sorting, the following method is used for processing:

[0031] When the contribution degree of the eigenvalue exceeds 1%, the current eigenvalue is selected as the feature quantity for subsequent sorting;

[0032] For the eigenvalues with contribution degrees between 0.5% and 1%, assuming there are e of them, calculate the weighted eigenvaluesθ is:

[0033] (19)

[0034] Reduce the eigenvalue with low contribution to 1;

[0035] For eigenvalues with a contribution lower than 0.5%, discard the corresponding feature quantities and consider them as noise;

[0036] After the above processing, if the number of selected feature quantities is p, it is considered that most of the information in the original data set can be covered;

[0037] Data set X p，m is used for subsequent sorting of batteries.

[0038] Furthermore, the preliminary sorting of S3 obtains the number of clusters and the cluster centers by using the data density. The specific process is as follows:

[0039] Step (1). First, according to the data set X composed of the extracted feature quantities p，m , calculate the data density range r for subsequent data density calculation, and calculate the maximum value of the distance between any two points:

[0040] (20)

[0041] In the formula, x i and x j represent the vectors corresponding to the data points of the i-th and j-th batteries, represents the 2-norm of the vector, that is, the square root of the sum of the squares of each element in the vector, and the superscript 2 represents taking the square of the calculation result.

[0042] To calculate the density of these points, define the influence range r of each point as:

[0043] (21)

[0044] In the formula, δ represents the desired degree of consistency, and the range is 0 to 1;

[0045] Step (2). Calculate the density of the i-th data point:

[0046] (22)

[0047] Step (3). Find the data point with the highest data point density as the first cluster center c 1 , then remove this data point and the influence range, and the points within the influence range calculated by formula (21). Then repeat steps (1) and (2), while looking for new clusters, dynamically adjust the data range;

[0048] Step (4). Loop sequentially until the new density is δ times the previous density, where δ 2 takes values from 0 to 1; 2 δ

[0049] Step (5). Obtain the initial value of the number of clusters and the initial positions of the cluster centers according to the classification situation at termination.

[0050] Furthermore, the further sorting in S4 is achieved by adjusting the distance from the data points to the cluster centers, and the specific process is as follows:

[0051] Step (1). Initialization: Use the number of clusters and the cluster centers obtained during preliminary sorting as the initialization parameters;

[0052] Step (2). Cluster the samples: Calculate the distance from each sample to each cluster center, and classify each sample into the class where the cluster center it is closest to is located:

[0053] (23)

[0054] In the formula, D ki represents the distance between the data point and the center point, c i represents the vector corresponding to the i-th center point, x j represents the vector corresponding to the data point corresponding to the j-th battery, represents the 2-norm of the vector, that is, the square root of the sum of the squares of each element in the vector.

[0055] Step (3). When the distance from the data point to two center points is equal, assign the point to the cluster with the larger current number of clusters; when the distance d ci from the point to each center point satisfies the following conditions:

[0056] (24)

[0057] That is, d ci is greater than half of the maximum distance between any two points in the data set, and the point is considered an outlier and used as a separate cluster center to accelerate the sorting speed; in the formula, x i and x j represent the vectors corresponding to the data of the i-th and j-th batteries.

[0058] Step (4). Calculate the new cluster centers: Calculate the central position of all data points in each cluster as the new cluster center c newi :

[0059] (25)

[0060] Where k is the number of data points in the cluster;

[0061] Step (5). When the change in the cluster center is small, the distance between two points is:

[0062] (26)

[0063] That is, when the distance between the two is less than the minimum distance between any two points in the data set, the clustering process is considered to end; otherwise, repeat steps (2) and (3);

[0064] Step (6). m batteries are sorted into g clusters, and the sorting is completed.

[0065] The rapid evaluation method for the consistency of power batteries provided by the present invention obtains eigenvalues by dimension reduction of the EIS curve data of the power batteries to be evaluated, pre-sorts the batteries using the data density, and then adjusts the cluster center by adjusting the distance between the data points and the cluster center, further sorting the batteries to improve the consistency of the batteries within the group after sorting.

[0066] Compared with the existing technology, the beneficial effects of the present invention are reflected in:

[0067] The technical solution of the present invention uses the alternating current impedance spectroscopy to test the battery and extracts the characteristic quantities of the battery from the EIS curve, which is faster than the conventional test method.

[0068] The technical solution of the present invention performs data dimension reduction on the EIS data, converts the consistency evaluation into a clustering problem of the dimension-reduced data, and reduces the influence brought by data redundancy.

[0069] The technical solution of the present invention sorts according to the data characteristics of the test data, does not depend on a specific battery model, is applicable to the sorting of various types of batteries, has a short test period and high working efficiency.

[0070] The technical solution of the present invention pre-sorts the batteries using the data density, and then adjusts the distance between the data points and the center point according to the obtained number of battery clusters and the center point, can adaptively select the number of clusters, and realizes the consistency sorting of power batteries with different quantity scales. Description of the Drawings

[0071] Figure 1 is a flowchart of a rapid evaluation method for the consistency of power batteries of the present invention;

[0072] Figure 2 is a flowchart of the data dimension reduction method in the embodiment;

[0073] Figure 3 is a flowchart of the battery clustering and sorting algorithm in the embodiment;

[0074] Figure 4 are the battery sorting results of the embodiments; among them, (a) shows the center points obtained by preliminary sorting, 5 center points are calculated, (b) shows the optimization of the center point positions after further sorting, (c) shows the preliminary sorting results, and (d) shows the results after further sorting. Specific Embodiments

[0075] To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0076] To improve the reliability and rapidity of the sorting of power battery consistency, the present invention proposes a method for rapid assessment of power battery consistency, which is a method for assessing power battery consistency based on the principal component analysis method and the clustering algorithm. The assessment method of the embodiments of the present invention will be described below with reference to the accompanying drawings:

[0077] As Figure 1 shown, the method for rapid sorting of power battery consistency in the embodiments of the present invention includes the following steps:

[0078] S1. Perform an alternating current impedance test on the battery pack to be tested to obtain a dataset of the electrochemical impedance spectrum (EIS) of the battery. The acquisition method is specifically:

[0079] Step (1): Apply excitation current signals with different frequencies to the battery. The excitation current value is taken as 1 / 20 of the rated capacity value of the battery, and synchronously sample the voltage signal U(t) and current signal I(t) of the battery;

[0080] Step (2): Through Fourier decomposition, extract the amplitudes of the battery voltage signal U(t) and current signal I(t) at frequency f;

[0081] (14)

[0082] In the formula, ω is the angular frequency corresponding to frequency f, are the phases corresponding to the voltage and current. n represents the nth injected frequency, and t represents the time interval;

[0083] Step (3): According to the battery voltage signal U(t) and current signal I(t), calculate the amplitude Z(n) and phase of the battery impedance under the excitation signal ;

[0084] (15)

[0085] Step (4): Change the frequency of the excitation signal, and repeat steps (2) and (3) to fit the electrochemical impedance spectrum curve of the battery through the impedance of the battery at different frequencies.

[0086] S2. Perform data dimensionality reduction on the sample data of the battery pack to be tested to obtain a new battery feature set;

[0087] Since the data obtained from the test is high-dimensional data containing a large amount of redundant information, which is not conducive to the sorting algorithm, a data dimensionality reduction method is used to extract features from the data. Figure 2 It is the flowchart of the data dimensionality reduction method in the embodiment, and the specific process is as follows:

[0088] The number of samples of the battery pack is m, and each EIS curve of the battery contains n points. Therefore, the original battery data is expressed as:

[0089] (16)

[0090] In the formula, X represents the set of all original battery data, and xm represents the column vector formed by the test data of the mth battery.

[0091] Construct the covariance matrix B of X n,m :

[0092] (17)

[0093] Perform eigenvalue decomposition on the covariance matrix B to find the eigenvalues λ i , and the corresponding eigenvectors v i . Arrange the eigenvectors v i in descending order by column from left to right according to the corresponding eigenvalue size to form a matrix, and take the first q columns to form a matrix T q，m .

[0094] The general definition of the contribution degree of eigenvalues is:

[0095] (18)

[0096] In the formula, λ i is the eigenvalue of the matrix T q，m .

[0097] In order to determine the number of selected feature quantities and prevent the generation of high-dimensional data from affecting subsequent sorting, the following method is used for processing:

[0098] When the contribution degree of the eigenvalue exceeds 1%, the current eigenvalue is selected as the feature quantity for subsequent sorting;

[0099] For eigenvalues with a contribution degree between 0.5% and 1%, assuming the number of them is e, calculate the weighted eigenvalue:

[0100] (19)

[0101] Reduce the eigenvalues with low contribution to 1

[0102] For eigenvalues with a contribution less than 0.5%, discard the corresponding feature quantities, which are considered to be noise

[0103] After the above processing, the number of selected feature quantities is p, which can be considered to cover most of the information of the original data set

[0104] Data set X p，m Will be used for subsequent sorting of batteries

[0105] S3. Calculate the data density of each point through the reduced data set, select the points with high density as the center points, and obtain the initial points of the sorting cluster centers and the number of clusters, including:

[0106] Use the reduced data to represent the battery to be tested and perform battery sorting Figure 3 It is the flow chart of the battery clustering and sorting algorithm in the embodiment. The clustering algorithm first pre-sorts the batteries based on the density of data points, and then performs further sorting. The specific method for sorting power batteries is as follows:

[0107] Since the data is in an unlabeled state during clustering, the battery sorting is transformed into an unsupervised clustering problem. Each data point is regarded as a potential cluster center. After subtracting the effect of the completed cluster center, the cluster center is searched again to provide the number of clusters and the initial value of the cluster center for subsequent further sorting. The specific process is as follows:

[0108] First, according to the data set X composed of the extracted feature quantities p，m , calculate the density range r of the data for subsequent calculation of data density, and calculate the maximum value of the distance between any two points:

[0109] (20)

[0110] In the formula, x i and x j represent the vectors corresponding to the data points of the i-th and j-th batteries, represents the 2-norm of the vector, that is, the square root of the sum of the squares of each element in the vector, and the superscript 2 represents taking the square of the calculation result

[0111] To calculate the density of these points, define the influence range of each point:

[0112] (21)

[0113] In the formula, δ represents the expected degree of consistency, with a range of 0 to 1. The smaller the value, the higher the degree of consistency, but the sorting speed will decrease accordingly.

[0114] Next, calculate the density of the i-th data point:

[0115] (22)

[0116] Find the data point with the highest density as the first clustering center c 1 , and then remove this data point and its influence range, as well as the points within the influence range calculated by formula (21). Recalculate the new influence range and density, and dynamically adjust the data range while searching for new clusters.

[0117] Then, loop sequentially until the new density is δ times the previous density 2 times, where δ 2 has a value range of 0 to 1. The smaller the value, the more classification categories will be generated, but the sorting speed will decrease. Generally, 0.5 is taken.

[0118] Finally, based on the classification situation at the end, obtain the initial value of the number of clusters and the initial positions of the clustering centers.

[0119] S4. Calculate the distance between each point and the center point, optimize the sorting result accordingly, and adjust the position of the clustering center and the batteries to be clustered, such as Figure 3 shown in the further sorting, including:

[0120] Since the center obtained from sorting is always a data point in the original dataset and may not necessarily be the most suitable clustering center, by adjusting according to the distance from the data point to the clustering center, a new clustering center is generated to achieve further sorting of the batteries. The specific process is as follows:

[0121] Initialization: Use the number of clusters and the set of initial clustering center points obtained during pre-sorting as initialization parameters;

[0122] Cluster the samples: Calculate the distance between each sample and each clustering center, and classify each sample into the class where the clustering center it is closest to is located.

[0123] (23)

[0124] In the formula, D ki represents the distance between the data point and the center point, c i represents the vector corresponding to the i-th center point, x j represents the vector corresponding to the data point corresponding to the j-th battery, represents the 2-norm of the vector, that is, the square root of the sum of the squares of each element in the vector.

[0125] Data point assignment: When the distances from a data point to two center points are equal, assign the point to the cluster with a larger current number of clusters; when the distance d from the point to each center point ci satisfies the following conditions:

[0126] (24)

[0127] That is, d ci is greater than half of the maximum distance between any two points in the data set, consider this point as an outlier, and use it as a cluster center to accelerate the sorting speed. Where x i and x j represent the corresponding vectors of the i-th and j-th battery data.

[0128] Calculate the new cluster center: Calculate the center position of all data points in each cluster as the new cluster center c newi .

[0129] (25)

[0130] In the formula, k is the number of data points contained in the cluster.

[0131] When the change in the distance between the old and new cluster centers is small:

[0132] (26)

[0133] That is, when the distance between the two is less than the minimum distance between any two points in the data set, the clustering algorithm stops; otherwise, continue to cluster the samples and calculate the new cluster center. Where c newi and c oldi represent the old and new center points respectively, D li represents the distance between the old and new center points, and d limit represents the minimum distance between any two points in the data set.

[0134] Based on this method, the sorting results were adjusted, improving the accuracy of clustering. As Figure 4 shown in the battery sorting results of a certain group of battery data, 3 feature quantities were extracted through data dimensionality reduction, and the corresponding 3D coordinates were plotted. Figure 4 (a) in Figure 4 shows the center points obtained by pre-sorting, and 5 center points were calculated. Figure 4 (b) in Figure 4 shows the optimization of the center point positions after further sorting. Figure 4 (c) in Figure 4Compared with (c) in it, outliers are effectively identified and classified separately, and the sorting results of clustering are also optimized.

[0135] After battery sorting using the clustering algorithm, the performance differences of the batteries sorted into the same cluster are small, and the batteries within the group have good consistency.

[0136] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the precision and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for rapid sorting of the consistency of power batteries, characterized in that, it includes the following steps: S1. For the battery samples to be measured, use the alternating current impedance method for testing, obtain the electrochemical impedance spectrum dataset of the batteries in the battery samples to be measured, and obtain the battery electrochemical impedance spectrum curve; S2. Perform data dimensionality reduction on the electrochemical impedance spectrum dataset of the battery samples to be measured, and use the data after dimensionality reduction to represent the batteries to be measured; the said data dimensionality reduction is to perform sorting and weighting processing on the eigenvalues based on eigenvalue decomposition to reduce the number of feature quantities. The specific process is as follows: The number of battery samples is m, and each battery is tested at n frequencies. The corresponding electrochemical impedance spectrum curve contains n points. Therefore, the original battery data is expressed as: (16) where X represents the set of all original battery data, and x m represents the column vector formed by the test data of the m-th battery; Construct X n,m Covariance matrix B of: (17) Perform an eigen - decomposition on the covariance matrix B to obtain the eigenvalues λ of the covariance matrix i , and the corresponding eigenvectors v i ; Arrange the eigenvector v i in descending order by column from left to right according to the corresponding eigenvalue magnitudes to form a matrix, and take the first q columns to form the matrix T q，m ; Contribution degree R of eigenvalue i The adopted definition is as follows: (18) where λ i is the eigenvalue of matrix T q，m ; In order to determine the selected feature quantities and prevent the generation of high-dimensional data from affecting subsequent sorting, the following processing method is adopted: When the contribution degree of the eigenvalue exceeds 1%, the current eigenvalue is selected as the feature quantity for subsequent sorting; For the eigenvalues with a contribution rate between 0.5% and 1%, assuming there are e of them, calculate the weighted eigenvalues θ as follows: (19) Reduce the eigenvalues with low contribution degrees to 1; For eigenvalues with a contribution degree lower than 0.5%, discard the corresponding feature quantities and consider them as noise; After the above processing, if the number of selected feature quantities is p, it is considered that most of the information of the original dataset can be covered; Dataset X p，m is used for the sorting of subsequent batteries; S3. Pre-sort the batteries based on the data after dimensionality reduction. According to the adaptive clustering algorithm, use the data density as an index to obtain the number of clusters and the cluster centers, and the density is adaptively updated as the iteration progresses; the pre-sorting uses the data density to obtain the number of clusters and the cluster centers. The specific process is as follows: Step (1). First, based on the dataset X composed of the extracted feature quantities p，m , calculate the density range r of the data for subsequent data density calculation, and calculate the maximum value of the distance between any two points: (20) where x i and x j represent the vectors corresponding to the data points of the i-th and j-th batteries, represents the 2-norm of the vector, that is, the square root of the sum of the squares of each element in the vector, and the superscript 2 represents squaring the calculation result; In order to calculate the density of these points, define the influence range r of each point as: (21) In the formula, δ represents the expected degree of consistency, and the range is 0 to 1; Step (2). Calculate the density of the i-th data point: （22） Step (3). Find the data point with the highest data point density as the first clustering center c 1 , and then remove this data point and its influence range, as well as the points within the influence range calculated by Equation (21). Repeat steps (1) and (2). While finding new clusters, dynamically adjust the data range; Step (4). Loop sequentially until the new density is δ times the previous density, where δ ranges from 0 to 1. 2 times, where δ 2 ranges from 0 to 1; Step (5). According to the classification situation at the end, obtain the initial value of the number of clusters and the initial position of the cluster centers; S4. Further sort the batteries. By calculating the distance between the data points and the cluster centers, adjust the cluster centers.

2. A method for rapid sorting of the consistency of power batteries according to claim 1, characterized in that, in the said S1, the specific process of obtaining the battery electrochemical impedance spectrum curve is as follows: Step (1). Apply excitation current signals with different frequencies to the batteries. The excitation current value is taken as 1 / 20 of the rated capacity value of the batteries, and synchronously sample the voltage signal U(t) and the current signal I(t) of the batteries; Step (2). Through Fourier decomposition, extract the amplitudes of the voltage signal U(t) and the current signal I(t) of the batteries at different frequencies; (14) where ω is the angular frequency corresponding to the frequency f, and are the phases corresponding to the voltage and current; n represents the n-th injected frequency, and t represents the time interval; Step (3). Calculate the amplitude Z(n) and phase of the battery impedance under the excitation signal based on the voltage signal U(t) and current signal I(t) of the battery : (15) Step (4). Change the frequency of the excitation signal, repeat steps (2) and (3), and fit the electrochemical impedance spectrum curve of the batteries through the impedance of the batteries at different frequencies.

3. A method for rapid sorting of the consistency of power batteries according to claim 1, characterized in that, the further sorting in the said S4 is achieved by adjusting the distance from the data points to the cluster centers. The specific process is as follows: Step (1). Initialization: Use the number of clusters and the cluster centers obtained during pre-sorting as the initialization parameters; Step (2). Cluster the samples: Calculate the distance from each sample to each cluster center, and classify each sample into the class where the cluster center it is closest to is located: (23) where D ki represents the distance between the data point and the center point, c i represents the vector corresponding to the i-th center point, x j represents the vector corresponding to the data point corresponding to the j-th battery, represents the 2-norm of the vector, that is, the square root of the sum of the squares of each element in the vector; Step (3). When the distances from a data point to two center points are equal, assign the point to the cluster with a larger current number of clusters; when the distance d ci satisfies the following condition: (24) That is, d ci is greater than half of the maximum distance between any two points in the data set. This point is considered an outlier and is used as a separate clustering center to accelerate the sorting speed. In the formula, x i and x j represent the vectors corresponding to the i-th and j-th battery data; Step (4). Calculate the new cluster centers: Calculate the central position of all data points in each cluster as the new cluster center c newi : (25) where k is the number of data points in the cluster; Step (5). When the change in the cluster center is small, the distance between two points is: (26) That is, when the distance between the two is less than the minimum distance between any two points in the data set, it is considered that the clustering process ends; otherwise, repeat steps (2) and (3); where c newi and c oldi represent the old and new central points respectively, D li represents the distance between the old and new central points, and d limit represents the minimum distance between any two points in the data set; Step (6). m batteries are sorted into g clusters, and the sorting is completed.

Images