A method and system for sorting liquid metal batteries

Abstract

The application discloses a sorting method and system of liquid metal batteries, and belongs to the technical field of secondary battery applications. The method comprises the following steps: reconstructing sampled liquid metal battery discharge voltage data to obtain a smooth voltage curve; identifying a voltage curve inflection point to generate a curve representation index; repeatedly mapping and screening the curve representation index to form a sorting index; filtering sample outliers in the sorting index by using an improved DBSCAN clustering algorithm; and then adopting a Mean Shift algorithm to optimally divide the remaining sample space to obtain a battery sorting result. The improved DBSCAN clustering algorithm is used to filter sample outliers, and then the Mean Shift algorithm is used to optimally divide the remaining sample space, so that the combined algorithm can simultaneously achieve the purposes of outlier detection and clustering, thereby improving the sorting precision.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of secondary battery application, and more particularly to a sorting method and system of liquid metal battery. BACKGROUND

[0002] As a new type of electrochemical energy storage technology, the liquid metal battery has the advantages of large capacity, high efficiency, long service life and low cost due to its three-layer liquid structure design, which can effectively solve the problems of short service life and low reliability faced by current electrochemical energy storage technology. In practical application, a large number of single batteries need to be connected in series and parallel to form a group to meet the voltage and capacity level requirements of large-scale energy storage systems. However, due to the differences in production and manufacturing processes, inconsistencies between single batteries are inevitable, specifically referring to the inconsistencies in battery internal resistance, capacity and other parameters. In addition, during the use of the battery group, the differences in application environment will further amplify the inconsistencies of each single battery, leading to rapid performance degradation and capacity utilization rate of the entire battery group. Even more likely, single battery failure will cause the entire battery group to malfunction, and in extreme cases, it may even cause safety accidents. Although theoretically, the liquid metal battery will not cause the safety accidents caused by the failure of lithium-ion batteries, the failure of single batteries will inevitably affect the utilization rate of the entire battery group system. Therefore, it is necessary to develop a battery sorting method for system application to realize scientific and efficient integration of batteries before grouping, which is conducive to long-term economic guarantee of the battery system.

[0003] A battery sorting method is disclosed in Chinese invention patent specification CN202110581420.3, which tests and sorts batteries according to a characteristic element, with fast sorting speed but low accuracy. A retired battery sorting method is disclosed in Chinese invention patent specification CN201811475223.8. The method obtains open-circuit voltage, constant-frequency resistance, discharge energy, direct-current resistance, and charge-discharge energy efficiency parameters of the retired battery through multiple test conditions, respectively, compares them with corresponding threshold values, and rejects the batteries not within the threshold range to realize battery sorting. The method considers five battery parameters comprehensively, with high sorting result reliability, but has many battery test conditions and long time consumption. A lithium-ion battery sorting parameter extraction method and battery sorting method are disclosed in Chinese invention patent specification CN202210768913.2. The method extracts three battery sorting parameters through lithium-ion battery electrochemical impedance spectrum test, relaxation time analysis, and grey correlation reconstruction, which can reflect battery dynamic changes and further improve battery sorting accuracy, but the parameter extraction process is complex, which may affect battery sorting efficiency. Therefore, while further constructing dynamic parameters reflecting battery state to improve classification accuracy, a simple and easy-to-operate method for quickly measuring or estimating dynamic parameters is needed to improve classification efficiency. In addition, the above methods are complete clustering, and when there are obvious outliers in the battery sorting parameters, the anti-interference ability of the sorting algorithm to the outliers is weak.

[0004] Due to the above defects and deficiencies, there is an urgent need in the art to make further improvements and refinements. According to the characteristics of liquid metal batteries, a battery parameter consistency screening method is developed, a rapid screening method based on the voltage curve of liquid metal batteries is designed, which effectively solves the problem of low battery sorting efficiency and improves the reliability of battery sorting, thereby ensuring the reliable and safe operation of the liquid metal battery pack. SUMMARY

[0005] In view of the defects of the prior art, the purpose of the present application is to provide a sorting method and system for liquid metal batteries, which aims to solve the problem of low efficiency of the existing battery sorting technology based on multiple parameters.

[0006] To achieve the above object, the application provides a quick battery sorting method based on liquid metal battery discharge curve, which forms battery sorting index by extracting feature parameters representing curve change from battery discharge curve, and simultaneously realizes battery outlier detection and adaptive spatial clustering division of sorting index by using improved DBSCAN (Density-Based Spatial Clustering of Applications with Noise) + Mean Shift (Mean Shift) combined clustering optimization algorithm, thereby improving battery sorting efficiency and accuracy.

[0007] To achieve the above object, the application provides a battery sorting method, which comprises the following steps:

[0008] The sampled liquid metal battery discharge voltage data is reconstructed to obtain a smooth voltage curve;

[0009] The voltage curve inflection point is identified to generate a curve representation index;

[0010] The curve representation index is repeatedly mapped and screened to form a sorting index;

[0011] The improved DBSCAN clustering algorithm is used to filter out sample outliers in the sorting index, and the Mean Shift algorithm is used to optimally divide the remaining sample space to obtain a battery sorting result.

[0012] The application first provides a battery discharge curve feature extraction framework, which extracts feature parameters that can effectively reflect the difference in curve shape on the discharge curve, eliminates the test process of obtaining battery parameters, and saves test time. The feature parameters can provide a reference for subsequent battery sorting.

[0013] Preferably, the feature extraction framework comprises four stages: data acquisition, data preprocessing, curve representation index generation, and battery sorting index generation.

[0014] The specific method is implemented according to the following steps:

[0015] (1) Collect the discharge data of the selected liquid metal battery at 0.2C rate, with a sampling time interval of 30s, to form an original data set;

[0016] (2) Smooth each group of data in the original data set;

[0017] Preferably, under the constant current discharge mode, the terminal voltage of the liquid metal battery should theoretically satisfy the following conditions:

[0018]

[0019] wherein is the terminal voltage of the battery at time t k k represents the kth sampling point.

[0020] Due to the influence of measurement device acquisition noise, the actual measured terminal voltage may not meet the condition of formula (1), and the curve needs to be smoothed by reconstructing the discharge voltage data. Preferably, the linear interpolation method is used to reconstruct the voltage curve, and the steps are as follows:

[0021] a. Mark the sample points: starting from the first sampling point, detection is carried out in turn. If a sample point meets the condition of formula (1), the sample point will be marked as a normal value (denoted as ), otherwise, it will be marked as an abnormal value (denoted as ), and the abnormal value must meet formula (2);

[0022]

[0023] b. Reconstruct the abnormal value: if the abnormal point satisfies formula (3), reconstruct the point (denoted as ) according to formula (4); if the abnormal point satisfies formula (5), reconstruct the j-1 points after the point according to formula (6), wherein the jth point satisfies formula (7):

[0024]

[0025]

[0026]

[0027]

[0028]

[0029] wherein is the terminal voltage of the battery at time t .

[0030] c. Update the abnormal point to a normal value in time after reconstruction, traverse all sampling points, and complete reconstruction.

[0031] (3) Identify the inflection point of the curve to generate a curve representation index;

[0032] The liquid metal battery discharge curve has two obvious voltage inflection points, and the battery voltage inflection point is one of the important characteristics reflecting the internal state of the battery, and therefore can be used to characterize the battery curve. Preferably, the present application uses a sliding window algorithm to identify the two curve inflection points, respectively. The voltage signal can be approximately regarded as linear change in a small time range, and therefore the present application uses an autoregressive model to model the voltage signal, and the related loss function is based on the least square residual:

[0033]

[0034] wherein is the voltage prediction value at time t. Assuming that the window width is 2ω, the loss function of the first window signal can be expressed as:

[0035]

[0036] The loss function of the second window is:

[0037]

[0038] The total loss function of the window is:

[0039]

[0040] The window signal difference value can be obtained according to formula (9), (10), (11):

[0041] Z(V tk-ω...tk+ω )=c(V tk-ω...tk+ω )-c(V tk-ω...tk )-c(V tk...tk+ω ) (12)

[0042] Preferably, the difference curve defined as Z high is obtained by identifying the first curve inflection point, and the difference curve defined as Z low is obtained by identifying the second curve inflection point. The time when the peak value of the difference curve appears corresponds to the time when the voltage inflection point appears, and therefore the time corresponding to the two inflection points can be obtained:

[0043]

[0044] wherein t high represents the time corresponding to the first inflection point, t low represents the time corresponding to the second inflection point, and the inflection point voltage value can be obtained:

[0045]

[0046] According to the voltage curve characteristics of the liquid metal battery, preferably, the present application extracts the following characteristic values as the curve characterization indexes:

[0047]

[0048]

[0049] Wherein F* represents the *th feature index, wherein F1-F4 represents two voltage inflection point positions, and F5-F7 can be used to represent the local shape of the curve.

[0050] (4) generating a sorting index.

[0051] Seven feature indexes are extracted herein. However, there may be repeated mapping information between different features, causing information redundancy, and thus affecting the effect of the sorting algorithm. Therefore, important features need to be selected from the above multi-dimensional features, so as to reduce the feature dimension and retain most of the effective information. Preferably, the present application uses a Spearman correlation coefficient matrix to select a new feature matrix with fewer features representing most of the information of the original feature matrix, thereby forming a sorting index. The steps are as follows:

[0052] a. Construct a multi-dimensional feature matrix A, A = [F1, F2, … Fn], n is the number of features, n = 7. Wherein F n is an m-row 1-column feature vector matrix, m is the number of battery samples; n

[0053] b. Sort the data in matrix A by column, and record the data position after sorting in matrix P, which is represented as:

[0054]

[0055] c. Calculate the correlation coefficient according to formula (18) to obtain the correlation between feature vectors:

[0056]

[0057] In the formula, ρ ij is the correlation coefficient between the i th feature vector and the j th feature vector.

[0058] d. Compare the correlation between the feature vectors, and take the feature vector with lower correlation as the sorting index (denoted as S), and the dimension of the sorting index is ≤3.

[0059] The present application also provides a sorting algorithm based on a combination algorithm.

[0060] Preferably, the present application uses an improved DBSCAN+Mean Shift algorithm to sort the battery, and the specific implementation steps are as follows:

[0061] ​(5) Standardization of sorting indicators: There are differences in dimensions and orders of magnitude among sorting indicators. To avoid the impact of these differences on the clustering results, the sample space of sorting indicators is first standardized. Preferably, the present invention uses zero-mean standardization, as shown in the following formula:

[0062]

[0063] In the formula, x represents the sorting index, and μ and σ are the mean and variance of each sorting index.

[0064] (6) Use the elbow detection method to determine the optimal DBSCAN algorithm neighborhood parameters (Eps, MinPts), where Eps is the neighborhood distance threshold and MinPts is the neighborhood sample number threshold. The initial value is the sorting index dimension plus 1.

[0065] 1) Calculate the k-distance. First, calculate the distance between each sample point and all other sample points. The formula is:

[0066]

[0067] Where r is the sorting index number and q is the sorting index dimension. Then... Sort the distances in ascending order. The sorted distance set is:

[0068] D(i)={d(i,1),d(i,2),…,d(i,k),…d(i,m)} (21)

[0069] Then d(i,k) is called The k-distance, where k = MinPts, and the initial value of MinPts is set to h+1 in this invention.

[0070] 2) Obtain the radius Eps: Plot all points in ascending order of their k-distance on the k-distance map, and take the point with the maximum curvature as Eps;

[0071] (7) DBSCAN algorithm clustering

[0072] 1) Initialization parameters: Core object collection Cluster number a = 0; Unvisited sample set H = E; Cluster partition set

[0073] 2) Traverse all sample points in set E to obtain the sample points. Eps-domain subset:

[0074]

[0075] Let b be the number of elements in this subset. If b ≥ MinPts, then... Add the core object set Ω;

[0076] 3) In the core object set Ω, randomly select a set element o, update the parameters:

[0077]

[0078] Based on this core object point, find the Eps-neighborhood sub-sample set N Eps (o) through neighborhood distance threshold, and put it into the temporary point set Z;

[0079] 4) Take a core object o' from the temporary point set Z, find all Eps-neighborhood sub-sample sets N Eps (o') through neighborhood distance threshold, and update

[0080]

[0081] 5) Repeat the above 4) process until the temporary point set Z = Φ, then the current clustering cluster P a is generated, and the cluster partition set is updated:

[0082] P = (P1, P2, …, P a ) (25)

[0083] Mark the non-core object points in the temporary point set Z as boundary points;

[0084] 6) Repeat the above 3)-5) process until the core object set is empty. The elements in the set (E-P) are marked as noise points, and the number of clustering clusters a is output;

[0085] 7) If a is 1, go to step 8), otherwise update MinPts and return to step (2), the update formula of MinPts is:

[0086] MinPts = MinPts + 1 (26)

[0087] 8) Output the first clustering result, and the number of samples in the set P is N;

[0088] (8) Mean Shift algorithm clustering

[0089] 1) Calculate the mean shift amount M h (x) of each sample point in the set P according to formula (21):

[0090]

[0091] Where K is the kernel function, and h is the bandwidth;

[0092] 2) Each sample point is translated, that is, moved along M​h (x) direction movement:

[0093] x i = x i + M h (x i ) (28)

[0094] 3) Repeat steps 1)-2) until the sample points converge, that is:

[0095] M h (x i ) = 0 (29)

[0096] 4) The sample points converging to the same point are considered to be the same cluster class.

[0097] 5) Output the second clustering result, that is, the battery sorting result.

[0098] The application also provides a sorting system for liquid metal batteries, comprising: a computer readable storage medium and a processor;

[0099] The computer readable storage medium is used to store executable instructions;

[0100] The processor is used to read the executable instructions stored in the computer readable storage medium, and execute the above-mentioned sorting method for liquid metal batteries.

[0101] Through the above technical solutions conceived by the application, compared with the prior art, the following beneficial results can be achieved:

[0102] 1. The application provides a method for extracting features of a liquid metal battery discharge curve. The voltage data is smoothed by a linear difference method to avoid the influence of collected noise on later feature extraction. The sliding window algorithm is used to identify the obvious voltage inflection point position of the liquid metal battery. Based on the voltage inflection point, 7 feature indexes that can be used to represent the battery discharge curve are extracted, and 3 key indexes are extracted through correlation analysis for battery sorting indexes. The index set can be directly applied to the subsequent battery sorting work.

[0103] 2. The application provides a combined sorting algorithm that can simultaneously realize outlier detection and battery clustering. The improved DBSCAN clustering algorithm is used to filter out sample outliers, and then the Mean Shift algorithm is used to divide the remaining sample space optimally. Therefore, the combined algorithm can simultaneously achieve the purposes of outlier detection and clustering, thereby improving the sorting accuracy.

[0104] 3. The quick battery screening method based on the voltage discharge curve provided by the application does not need to make accurate measurement on the battery parameters, and only needs to rely on the battery discharge curve, so that the liquid metal battery can be quickly sorted, the operation is simple, the test time is short, and the data processing is convenient. In combination with the characteristics of the liquid metal battery and the application environment characteristics, the method is a quick and reliable battery sorting method. BRIEF DESCRIPTION OF DRAWINGS

[0105] Figure 1 A flow chart of the construction method of the battery sorting based on the discharge voltage curve provided by the application is provided.

[0106] Figure 2 A liquid metal battery voltage curve provided by the embodiment of the application is provided, wherein (a) is an original voltage curve of a liquid metal battery and a voltage curve after data reconstruction thereof; and (b) is 212 voltage curves after data reconstruction.

[0107] Figure 3 A curve inflection point identification result diagram provided by the embodiment of the application is provided, wherein (a) is an identification result of the first inflection point of the discharge curve; and (b) is an identification result of the second inflection point of the discharge curve.

[0108] Figure 4 A feature index Spearman correlation coefficient matrix provided by the embodiment of the application is provided.

[0109] Figure 5 A distribution diagram of each sorting index and a sample space diagram formed by the same provided by the embodiment of the application are provided, wherein (a), (b) and (c) are respectively scatter distribution diagrams of the sorting index S1, the sorting index S2 and the sorting index S3 after standardization processing; and (d) is a sample space formed by the above three sorting indexes.

[0110] Figure 6 An algorithm clustering result provided by the embodiment of the application is provided, wherein (a) is a DBSCAN algorithm output clustering result; and (b) is a Mean Shift algorithm output clustering result.

[0111] Figure 7 A liquid metal battery sorting result diagram provided by the embodiment of the application is provided, wherein (a) is a discharge curve diagram of an outlier single battery after sorting; (b) is a battery voltage curve corresponding to the cluster C1; (c) is a battery voltage curve corresponding to the cluster C2; (d) is a battery voltage curve corresponding to the cluster C3; (e) is a battery voltage curve corresponding to the cluster C4; and (f) is a battery voltage curve corresponding to the cluster C5. DETAILED DESCRIPTION

[0112] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application 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 application and should not be used to limit the present application. In addition, the technical features involved in the various embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0113] The present application provides a sorting method of liquid metal battery, comprising the following steps:

[0114] The sampled liquid metal battery discharge voltage data is reconstructed to obtain a smooth voltage curve;

[0115] The inflection point of the voltage curve is identified to generate a curve representation index;

[0116] The curve representation index is repeatedly mapped and screened to form a sorting index;

[0117] The improved DBSCAN clustering algorithm is used to filter out sample outliers in the sorting index, and the Mean Shift algorithm is used to optimally divide the remaining sample space to obtain a battery sorting result.

[0118] In an embodiment of the present application, the battery used is a 200Ah Li||Bi liquid metal battery prepared in the laboratory.

[0119] Figure 1 is a flowchart of the battery sorting method provided by the present application, and the main steps thereof include:

[0120] (1) obtaining battery data;

[0121] After the activation of the liquid metal battery is completed, only the discharge curve of each liquid metal battery at 0.2C rate needs to be tested to obtain the original data set. In this embodiment, a total of 212 200Ah Li||Bi liquid metal battery voltage curves were measured;

[0122] (2) smoothing the discharge data and identifying the inflection point of the curve;

[0123] (3) generating a sorting index set;

[0124] (4) optimizing the neighborhood parameter of the DBSCAN algorithm, using the DBSCAN algorithm to cluster the sample space formed by the sorting index set, and marking the outlier sample points;

[0125] (5) removing the outlier sample points, and using the Mean Shift algorithm to re-cluster the sample space to obtain a battery sorting result.

[0126] Figure 2(a) in FIG. 1 is the original voltage curve of a liquid metal battery provided in the embodiment and the voltage curve after data reconstruction thereof, and Table 1 gives the relevant information of the first 6 sampling points in the enlarged view, including 3 normal sampling points, 3 abnormal sampling points and the corresponding reconstructed values. The first two abnormal values meet the condition of formula (5), and the reconstructed values thereof are processed according to formula (7); the third abnormal value meets the condition of formula (3), and the reconstructed value thereof is processed according to formula (4). Figure 2 (b) in FIG. 1 is a discharge voltage curve diagram of 212 liquid metal batteries provided in the embodiment after smoothing processing. As can be seen from the figure, the curve shapes of the batteries are obviously different, reflecting that the difference of the battery parameters is large.

[0127] Table 1: Reconstructed values of sampling point voltages

[0128]

[0129] Figure 3 is the curve inflection point recognition result figure provided in the embodiment, and as can be seen from the figure, the sliding window algorithm used can accurately identify the voltage inflection point position. Based on the voltage inflection point position, a characteristic index representing the curve shape can be generated.

[0130] Figure 4 is the characteristic index Spearman correlation coefficient matrix provided in the embodiment. The correlation coefficient matrix is a symmetric matrix, and the rows and columns describe the same series of characteristic quantities. The correlation of one or more characteristics is high, indicating that they have similar trends and may contain similar information. In general, when the correlation coefficient is less than 0.20, it is considered that the variables are weakly correlated or not correlated. If there are more than 3 correlation coefficients less than 0.20 (50% of the number of characteristic quantities) in a column, it is considered that the characteristic has low correlation and is an independent characteristic. As can be seen from the figure, F3 and F4 are independent characteristics, F1 and F5 have low correlation with them, and F5 has lower correlation. However, F1 and F5 have strong correlation with themselves. According to the above analysis, the final characteristic dimension is determined to be 3 dimensions, and thus the dimension reduction selection of the multi-dimensional characteristic quantity is completed, and the corresponding characteristics are F3, F4 and F5. Then, the characteristics F3, F4 and F5 are selected as the final sorting indicators and are marked as S1, S2 and S3 respectively.

[0131] Figure 5 is the sample space figure constituted by the distribution figures of the respective sorting indicators provided in the embodiment. Figure 5 (a), (b) and (c) in FIG. 1 are respectively the scatter distribution figures of the sorting indicator S1, the sorting indicator S2 and the sorting indicator S3 after standardization processing, and (d) is the sample space constituted by the above three sorting indicators, and there are 212 sample points. The sample distribution has obvious high-density clustering center, and the density distribution conforms to the rule that the middle is large and the edge is small.

[0132] Figure 6 is the clustering result of the algorithm provided by the present application. The optimized neighborhood parameter is (1.152, 7), and the final DBSCAN algorithm output clustering result is shown in (a) of FIG. 6, and the statistical result is shown in Table 2. Figure 6 Figure 6 From (a) of FIG. 6 and Table 2, it can be seen that the DBSCAN algorithm generates a cluster, which contains 190 sample points, indicating that 190 single batteries are sorted out, indicating that this part of the single battery has relatively good consistency, of which there are 137 core sample points, and this part of the single battery has better consistency. The algorithm marks 22 outliers as noise points, indicating that there are 22 outlier single batteries in the embodiment. The clustering result of the Mean Shift algorithm is shown in (b) of FIG. 6, and the algorithm divides the sample space composed of 190 sample points into 5 clusters, and the number of samples in the cluster is 101, 41, 20, 15 and 13 respectively. Table 3 is the clustering center of each cluster, and the indicators of cluster C1 are closest to the average level, and the number of samples in this group is also the most, indicating that the samples have obvious clustering form; the second inflection point voltage (indicator S2) of cluster C3 is lower than that of cluster C2, and the other two indicators are similar and the value of indicator S3 is relatively high; the second inflection point voltage of cluster C4 is the highest and the corresponding time (indicator S1) is the longest, in addition, the value of indicator S3 is also the lowest. The second inflection point voltage of cluster C5 is the lowest and the corresponding time is the shortest. Figure 6

[0133] Table 2 DBSCAN clustering algorithm output result

[0134]

[0135] Table 3 Clustering cluster center

[0136]

[0137] Figure 7 is the battery sorting result schematic diagram provided by the present application. Figure 7 (a) of FIG. 6 is the voltage curve corresponding to all sample batteries, in which the dashed line represents the curve of the battery corresponding to the noise point, and from the figure it can be seen that the discharge curve of the battery corresponding to the noise point is distributed at the edge position of the whole curve cluster. In addition, the battery voltage curves corresponding to each cluster are respectively shown in (b) and (c) of FIG. 6. Figure 7 ​​(b), (c), (d), (e) and (f) in the figure. It can be seen from the figure that the voltage curves in each cluster have high coincidence. It can be seen that the method provided by the application improves the consistency of the batteries significantly based on the feature extraction and clustering analysis method according to the inconsistency information reflected by the 212 battery monomers on the discharge curve, and the 22 batteries with large differences are excluded, and then the clustering is performed. The results show that the method has strong feasibility, and can quickly and reliably screen out abnormal batteries in actual application, thereby improving the consistency of the grouped batteries.

[0138] Those skilled in the art will easily understand that the above description is only the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for sorting liquid metal batteries, characterized in that, Includes the following steps: The sampled liquid metal battery discharge voltage data were reconstructed using a linear interpolation method to obtain a smooth voltage curve; A sliding window algorithm is used to identify inflection points on the voltage curve and generate curve characterization indicators, including: Voltage is modeled using an autoregressive model, with the relevant loss function based on least-squares residuals: ,in for t The predicted voltage value at time 2; assuming a window width of 2. ω The loss function for the first window is then expressed as: The loss function for the second window is expressed as: The total window loss function is , obtain window difference ; The times corresponding to the two inflection points are: ,in To identify the difference curve obtained from the first inflection point of the curve, The difference curve obtained to identify the inflection point of the second curve; Two inflection point voltage values ​​were obtained: ,in t high This indicates the time corresponding to the first inflection point. t low This indicates the time corresponding to the second inflection point; Repeated mapping and filtering of the curve characterization indicators are performed to form sorting indicators; The improved DBSCAN clustering algorithm is used to filter out outliers in the sorting criteria, and then the Mean Shift algorithm is used to optimally partition the remaining sample space to obtain the battery sorting results. In constant current discharge mode, the terminal voltage of a liquid metal battery satisfies the following condition: in That is, the battery is The terminal voltage at time t, where k Indicates the first k One sampling point.

2. The method according to claim 1, characterized in that, The voltage curve is reconstructed using a linear interpolation method, and the steps include: Starting from the first sampling point, the tests are performed sequentially. If a sampling point meets the requirements... If the sample point is not normal, it will be marked as a normal value, denoted as . Otherwise, it will be marked as an outlier, denoted as And outliers must meet the following criteria: ; If the outlier satisfies Then according to Reconstruct this point, denoted as If the anomaly satisfies Then according to Reconstruction begins from that point onwards. j -1 points, where the first point is... j All points meet the conditions. , That is, the battery terminal voltage is The corresponding moment; After reconstruction, outliers are updated to normal values ​​in a timely manner. The reconstruction is completed by traversing all sampling points.

3. The method according to claim 1, characterized in that, Curve characterization metrics include: Where F* represents the *th characteristic index, F1-F4 characterize the positions of two voltage inflection points, and F5-F7 can be used to characterize the local shape of the curve.

4. The method according to claim 3, characterized in that, The sorting indicators specifically include: Constructing a multidimensional eigenvalue matrix A , A =[ F 1, F 2,… F 7], of which F n for m The eigenvector matrix with row 1 and column 1, m Number of battery samples n The number of features; For matrix A The data in the matrix is ​​sorted by column. P Record the position of the sorted data, denoted as: ; The correlation coefficient is calculated to obtain the correlation between feature vectors: In the formula ρ ij For the first i The eigenvector and the eigenvector j The correlation coefficient between the feature vectors; Compare the correlation between feature vectors and take the feature vector with lower correlation as the sorting index. The dimension of the sorting index is ≤3.

5. The method according to claim 1, characterized in that, The process involves using an improved DBSCAN clustering algorithm to filter out outliers in the sorting criteria, followed by employing the Mean Shift algorithm to optimally partition the remaining sample space to obtain the battery sorting results. Specifically, this includes: (1) Zero-mean standardization was used for the sorting indicators: In the formula x σ represents the sorting index, where μ and σ are the mean and variance of each sorting index. (2) Determine the optimal neighborhood parameters of the DBSCAN algorithm using the elbow detection method. Eps and MinPts ,in Eps It is the neighborhood distance threshold. MinPts It is the threshold for the number of neighborhood samples, and the initial value is taken as the sorting index dimension plus 1; (3) DBSCAN algorithm clustering (31) Initialization parameters: Core object collection Number of clusters Unvisited sample set Cluster partitioning set ; (32) Traversing a set E All sample points are obtained. of Eps -Domain Subsample Set: The number of elements in this subset is denoted as ,like b ≥ MinPts Then Add to core object collection middle; (33) In the core object set In the middle, randomly select an element from the set. Update parameters: Based on this core object point, the neighborhood distance threshold is used to find... Eps - Neighborhood Subsample Set Place it into a temporary point set middle; (34) From the temporary point set Extract a core object Find all of them by using the neighborhood distance threshold. Eps - Neighborhood Subsample Set Update parameters: ; (35) Repeat the above process (34) until a temporary point set is reached. Then the current cluster Once generated, update the cluster partition set: Mark the non-core object points in the temporary point set Z as boundary points; (36) Repeat steps (33)-(35) above until the core object set is empty, where the set The elements in the data are marked as noise points, and the number of clusters is output. a ; (37) If a If the value is 1, proceed to step (38); otherwise, update. MinPts Return to step (32). MinPts The update formula is: ; (38) Output the first clustering result, and denote the number of samples in set P as . N ; (4) Mean Shift algorithm clustering (41) Calculate the mean shift for each sample point in set P. : ;in For kernel function, h For bandwidth; (42) Each sample point is translated, that is, along... Directional movement: ; (43) Repeat steps (41)-(42) until the sample points converge, i.e.: ; (44) Sample points that converge to the same point are considered to belong to the same cluster. (45) Output the second clustering results, i.e. the battery sorting results.

6. The method according to claim 5, characterized in that, The method of determining the optimal neighborhood parameters of the DBSCAN algorithm using elbow detection includes: Calculating k-distance: First, calculate the distance between each sample point and all other sample points. The formula is: Then Sort the distances in ascending order, the sorted distance set is: ,but It is called The k-distance, where r For sorting index serial numbers, q As for the sorting index dimensions, k = MinPts ; Get radius Eps Plot the k-distances of all points in ascending order on a k-distance map. Eps Take the point of maximum curvature.

7. A sorting system for liquid metal batteries, characterized in that, include: Computer-readable storage media and processors; The computer-readable storage medium is used to store executable instructions; The processor is used to read executable instructions stored in the computer-readable storage medium and execute the sorting method of liquid metal batteries according to any one of claims 1 to 6.

Images