Secondary screening method for decommissioned batteries connected in parallel to form groups

Abstract

The invention relates to the technical field of echelon utilization of retired batteries, in particular to a secondary screening method for parallel grouping of retired batteries, which comprises the following steps: extracting the maximum available capacity, ohmic internal resistance, charge transfer internal resistance, diffusion impedance and average voltage of a single battery as multi-dimensional screening characteristics, and comprehensively representing the health state of the battery; the high-dimensional feature matrix is reduced to a two-dimensional space by using a t-SNE algorithm, and the nonlinear relation is reserved; first-level sorting is achieved in combination with DBSCAN density clustering, noise is removed, and grouping is carried out; pre-assembling batteries in the same cluster, dynamically monitoring the temperature difference in the group, and setting an inconsistency evaluation threshold for secondary verification; and if not, positioning and replacing an abnormal battery based on a Z-Score method, and iteratively optimizing the matching sequence. Through multi-parameter electrochemical characteristic fusion and a dynamic and static two-stage verification mechanism, the dynamic consistency of the battery pack is improved, abnormal accurate replacement is supported, the recombination success rate is improved, the echelon utilization life is remarkably prolonged, and the safety risk is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of cascade utilization of retired batteries, and in particular to a secondary screening method for grouping retired batteries in parallel. Background Art

[0002] As the global energy transition accelerates, the large-scale application of lithium-ion batteries in electric vehicles and energy storage systems has created a massive demand for recycling retired batteries. According to industry forecasts, by 2030, the total amount of retired power batteries worldwide will exceed 5 million tons, of which more than 70% have the potential for cascade utilization. However, due to the uneven performance degradation and complex internal aging mechanisms of retired batteries caused by long-term recycling, their efficient sorting and reassembly have become the core bottleneck restricting the large-scale development of cascade utilization. Current technologies face the following key challenges:

[0003] Existing screening methods rely on single parameters and cannot fully characterize the battery's state of health (SOH). Traditional methods often sort based on capacity or ohmic internal resistance, ignoring the multidimensional nature of the battery's internal electrochemical properties. For example, batteries with similar capacities may exhibit significant differences in charge transfer impedance or diffusion impedance, leading to mismatched dynamic responses after grouping. High-dimensional data processing is inefficient and subject to severe noise interference. The "curse of dimensionality": As the dimension of battery features increases, traditional linear dimensionality reduction methods are unable to capture nonlinear relationships, resulting in decreased clustering accuracy. Furthermore, algorithms such as K-means require a preset number of clusters and are sensitive to noise. A dynamic anomaly detection mechanism is lacking before retired batteries are grouped, resulting in high costs for failed reorganizations. Existing technologies lack real-time monitoring and precise location of dynamic anomalies after grouping, often employing a strategy of eliminating entire groups, resulting in significant resource waste. An innovative approach combining multi-dimensional feature fusion, data-driven sorting, and dynamic verification iteration is urgently needed to overcome the bottleneck of efficient reorganization of retired batteries.

[0004] To address the above problems, the present invention constructs a full-chain sorting system by integrating electrochemical impedance spectroscopy, nonlinear dimensionality reduction, density clustering, pre-assembly testing and Z-Score anomaly positioning, providing reliable technical support for large-scale cascade utilization. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention proposes a two-level screening method for retired batteries in parallel groups, which can improve the dynamic consistency of the battery pack, support abnormal and accurate replacement, improve the success rate of reorganization, significantly extend the cascade utilization life and reduce safety risks.

[0006] The technical solution adopted by the present invention is: a two-stage screening method for retired batteries connected in parallel, comprising the following steps:

[0007] Step S1: Perform performance tests on a batch of batteries with different health states, select and extract several items of internal electrochemical characteristic data of each battery, and use the extracted high-dimensional feature values after data standardization as screening features;

[0008] Step S2: Mapping the high-dimensional feature data extracted from each single battery into a two-dimensional space using the t-SNE algorithm to obtain the two-dimensional features of each single battery;

[0009] Step S3: Based on the two-dimensional features of each single battery obtained in step S2, the DBSCAN method is used to perform first-level cluster sorting on the batch of batteries, so as to obtain the battery groups of the same cluster with similar first-level cluster sorting positions, and eliminate the noise batteries with discrete first-level cluster sorting positions;

[0010] Step S4: pre-assemble the same cluster batteries selected in step S3 using a parallel device, and test the inconsistency of these batteries after parallel grouping for secondary inspection. Batches of batteries that meet the inconsistency requirements can be parallel grouped and used as needed;

[0011] Step S5: For pre-assembled battery packs that fail the secondary test, the Z-Score method is used to locate mismatched batteries, and the abnormal batteries are replaced with the same cluster batteries screened in step S3. The inconsistency test in step S4 is continued until a batch of batteries with qualified parallel matching are screened out.

[0012] Furthermore, in step S1, the extracted internal electrochemical characteristic data of each battery are the maximum available capacity, ohmic impedance, charge transfer impedance, diffusion impedance and average voltage. The maximum available capacity of the battery is the capacity released during the process of discharging the battery at a constant current of 1C to the cut-off voltage when the battery is fully charged; the ohmic internal resistance, charge transfer internal resistance and diffusion impedance are obtained by fitting the electrochemical impedance spectroscopy (EIS) with a first-order RC model. Furthermore, the EIS curve is tested when the battery is at 50% SOC, and the scanning frequency is 0.005Hz~100KHz; the average voltage is the average voltage of the battery during the plateau period corresponding to 20%-80% SOC during the 1C constant current discharge process.

[0013] As a preferred solution, in step S1, the data normalization method selected is Z-Score normalization, and its core formula is:

[0014] Where X is the original data point; u is the mean of the data set; and σ is the standard deviation of the data set.

[0015] As a preferred solution, the core steps of using the t-SNE algorithm to map the high-dimensional data feature matrix to a two-dimensional space in step S2 are as follows:

[0016] Step S2.1, calculating high-dimensional space similarity;

[0017] For a high-dimensional data point x i and x j , calculate x by the formula j As x i The conditional probability of the neighbor, reflecting x i The similarity with surrounding points is:

[0018] Where σ i Represents the perplexity, controlling the number of effective neighbors for each point, where x i 、x j Represent the high-dimensional data features of different batteries labeled i and j respectively;

[0019] Step S2.2, constructing a low-dimensional space probability distribution;

[0020] In low-dimensional space, use the t distribution with 1 degree of freedom to calculate the low-dimensional data point y i and y j Similarity, the formula is:

[0021] Where y i 、y j , yk, yl are low-dimensional data points corresponding to batteries with different labels mapped to low-dimensional data;

[0022] Step S2.3, minimizing KL divergence based on gradient descent to preserve the nonlinear relationship of the data;

[0023] Calculate the KL divergence between the high-dimensional probability distribution P and the low-dimensional probability distribution Q. By minimizing the KL divergence, the low-dimensional space can reproduce the local structure of the high-dimensional space as much as possible. The smaller the KL divergence, the smaller the difference between the high-dimensional and low-dimensional distributions. The formula is:

[0024] Based on the gradient descent method, the low-dimensional coordinates are optimized by calculating the KL divergence of the low-dimensional coordinates y i The gradient of the data point in the low-dimensional space is iteratively updated, and the low-dimensional distribution is gradually optimized to make it close to the high-dimensional distribution. The formula is:

[0025] As a preferred solution, the perplexity parameter of the t-SNE algorithm in step S2 is set to 2-5, and the neighborhood radius of the DBSCAN algorithm in step S3 is set to 0.05-0.08. The batteries within the neighborhood radius are marked as part of the same cluster battery group, and the batteries outside the neighborhood radius are marked as noise batteries.

[0026] As a preferred solution, in step S4, the inconsistency test condition after pre-assembly of the battery is to discharge the fully charged parallel battery pack at a constant current of 1C to the cut-off voltage at 25°C; the temperature data of each branch battery during the discharge process is collected by a temperature sensor, and the inconsistency of the battery pack is described by the difference in temperature distribution. The specific inconsistency evaluation method is as follows:

[0027] Step S4.1: calculate the instantaneous maximum temperature difference T F Determined as the inconsistency assessment indicator, the formula is as follows: T max_i =max{T 1_i ,T 2_i ...T n_i},i=1,2,3...k T min_i =min{T 1_i ,T 2_i ...T n_i},i=1,2,3...k ΔT max =max{T max_i -T min_i},i=1,2,3...k

[0028] Where, T n-i is the temperature value of the nth temperature sensor at time i, and the total time is k; T max-i and T min-i represents the maximum and minimum temperatures within the group at time i, ΔT max It is the instantaneous maximum temperature difference within the battery pack during the entire discharge process;

[0029] Step S4.2: Set the inconsistency evaluation index ΔT max The threshold T h The regulation is 5℃;

[0030] Step S4.3, calculate the inconsistency score G of the pre-assembled battery pack, with a score range of [0-100], and the formula is:

[0031] The qualification condition for pre-assembled battery packs is that the inconsistency score G≤50. A qualified battery pack indicates that there is good matching between the pre-assembled single cells, among which G≤25 indicates that the battery pack has excellent consistency and can be used in high-requirement equipment.

[0032] Furthermore, in step S5, the abnormal battery is located by using the Z-Score method to calculate the degree to which the average temperature of each battery deviates from the average temperature of other batteries. The Z-Score value of the branch battery is calculated based on the temperature data of each battery in the group obtained by the secondary detection. The maximum value of |Z-Score| is the corresponding abnormal battery in the pre-assembled battery pack. The formula is:

[0033] Where, is the average temperature of the jth battery; u group ,σ group are the mean and standard deviation of temperature within the group, respectively.

[0034] Furthermore, if the iterative optimization in step S5 fails to meet the requirements after more than three times, the entire battery pack is downgraded for use in low-requirement scenarios.

[0035] The beneficial effects of the present invention are:

[0036] First, the present invention improves screening accuracy through multi-parameter collaborative optimization. It introduces multi-dimensional impedance parameters such as ohmic impedance, charge transfer impedance, and diffusion impedance, and combines the capacity and voltage platform characteristics to accurately identify the battery health status characteristics;

[0037] Second, the present invention achieves a balance between efficiency and reliability through data-driven sorting. t-SN E maps high-dimensional features into a two-dimensional space, preserving nonlinear relationships, avoiding the "curse of dimensionality," and significantly improving computational efficiency. It also utilizes DBSCAN for cluster sorting, adaptively identifying battery clusters with similar parameters and eliminating noise points.

[0038] Third, the present invention ensures group consistency through two-level dynamic verification. Pre-assembly testing simulates actual operating conditions, monitoring the dynamic temperature changes of each battery in the group during constant current discharge to quantify inconsistency scores. Z-Score anomalies are detected for unqualified pre-assembled battery packs, allowing accurate replacement of abnormally heated batteries to avoid entire packs from being scrapped and improve reassembly success rates. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 This is a flow chart of the secondary screening method for retired batteries connected in parallel groups according to the present invention;

[0041] Figure 2 The distribution diagram of the five standardized features extracted from each single battery;

[0042] Figure 3 This is the DBSCAN first-level clustering sorting diagram after t-SNE dimension reduction;

[0043] Figure 4 Temperature distribution and maximum temperature difference curves within the pre-assembled battery packs: (a) pack one; (b) pack two; (c) pack three; (d) pack four. DETAILED DESCRIPTION

[0044] The present invention is described in detail below by way of exemplary embodiments, but it should be understood that elements, structures, and features of one embodiment may be beneficially combined in other embodiments without further description.

[0045] It should be noted that, unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning understood by persons having ordinary skills in the field to which the invention belongs. The words "one," "an," "the," and similar expressions used in the patent specification and claims of the present invention do not express quantitative limitations, but rather indicate the presence of at least one; words such as "include" or "comprise" indicate that the elements or objects preceding "include" or "comprise" include the elements or objects listed after "include" or "comprise" and their equivalents, but do not exclude other elements or objects with the same function.

[0046] In order to more clearly describe the method of the present invention, Figure 1 -Attached Figure 4 Describe this embodiment:

[0047] Figure 1 This is a flow chart of a secondary screening method for retired batteries in parallel groups according to the present invention, which specifically includes the following steps:

[0048] Step 1: A batch of 32 batteries with a health status of 70%-100% is available. Performance tests are performed on these batteries to extract the maximum available capacity C, ohmic resistance R0, charge transfer resistance Rc, diffusion resistance Rw, and average voltage U of each battery.

[0049] In this embodiment, the maximum available capacity of the battery is the capacity released during the process of discharging the battery at a constant current of 1C to the cut-off voltage when the battery is fully charged; the ohmic internal resistance, charge transfer internal resistance, and diffusion impedance are obtained by fitting the electrochemical impedance spectroscopy (EIS) with a first-order RC model. Furthermore, the EIS curve is tested when the battery is at 50% SOC, with a scanning frequency of 0.005Hz to 100kHz; the average voltage is the average voltage of the battery during the plateau period corresponding to 20%-80% SOC during the 1C constant current discharge process;

[0050] The data is normalized by Z-Score and used as a screening feature to eliminate dimensional differences and balance feature weights. The core formula is:

[0051] Where X is the original data point; u is the mean of the data set; σ is the standard deviation of the data set, Figure 2 The distribution of the five standardized features extracted from each single battery is shown;

[0052] Step 2: Currently difficult to Figure 2 The five features shown are used to directly sort the batteries, so the t-SNE algorithm is used to map the high-dimensional feature data extracted from each single battery into a two-dimensional space.

[0053] The core steps of using the t-SNE algorithm to map the high-dimensional data feature matrix to a two-dimensional space in step 2 are as follows:

[0054] Step 2-1 calculates the high-dimensional space similarity;

[0055] For a high-dimensional data point x i and x j , calculate x by the formula j As x i The conditional probability of the neighbor, reflecting x i Similarity to surrounding points.

[0056] Where: σ i Represents the perplexity, which controls the number of effective neighbors of each point, where Xi and Xj represent the high-dimensional data features of different batteries labeled i and j respectively.

[0057] Step 2-2 constructs a low-dimensional space probability distribution;

[0058] In low-dimensional space, use the t distribution with 1 degree of freedom to calculate the low-dimensional data point y i and y j Similarity.

[0059] where y i 、y j 、y k 、y l The high-dimensional data corresponding to batteries with different labels are mapped to low-dimensional data points of low-dimensional data.

[0060] Steps 2-3 minimize the KL divergence based on the gradient descent method to preserve the nonlinear relationship of the data.

[0061] Calculate the KL divergence of the high-dimensional probability distribution P and the low-dimensional probability distribution Q. By minimizing the KL divergence, the low-dimensional space can reproduce the local structure of the high-dimensional space as much as possible. The smaller the KL divergence, the smaller the difference between the high-dimensional and low-dimensional distributions. ij With q ij They are the conditional probabilities of the high-dimensional space and low-dimensional space calculated above:

[0062] Based on the gradient descent method, the low-dimensional coordinates are optimized by calculating the KL divergence of the low-dimensional coordinates y i The gradient of , iteratively updates the coordinates of the data points in the low-dimensional space, and gradually optimizes the low-dimensional distribution to make it close to the high-dimensional distribution.

[0063] In this embodiment, the perplexity is set to 2;

[0064] Step 3: Based on the two-dimensional features of each single battery obtained in step 2, the DBSCAN method is used to perform first-level cluster sorting on the batch of batteries, where the neighborhood radius is set to 0.06 in this embodiment. The cluster sorting results are as follows: Figure 3 As shown, the grouping is as follows: Cluster 1: #2, #3, #11, #13; Cluster 2: #19, #21, #23, #24, #25, #26, #30; Cluster 3: #1, #4, #7, #27, #29, #31, #32;

[0065] Among them, batteries numbered #5, #6, #8, #9, #10, #12, #14, #15, #16, #17, #18, #20, #22 and #28 are marked as noise points. However, this does not mean that these batteries have no use value. It is just that due to the limitation of sample quantity, batteries suitable for parallel connection have not been matched yet.

[0066] Step 4: Select 3 batteries from each of the 3 battery clusters selected in step 3. The selection of the 3 battery clusters is as follows: Cluster 1: #3, #11, and #13 (denoted as group 1); Cluster 2: #19, #21, and #23 (denoted as group 2); Cluster 3: #1, #4, and #7 (denoted as group three);

[0067] Pre-assemble with a detachable parallel device and test the temperature inconsistency of these batteries after parallel connection for secondary inspection;

[0068] In step 4, the inconsistency test condition after pre-assembly of the battery is to discharge the fully charged parallel battery pack at a constant current of 1C to the cut-off voltage at 25°C;

[0069] Temperature sensors are used to collect temperature data from each battery branch during discharge. The inconsistency of the battery pack is described by the difference in temperature distribution. The specific inconsistency assessment method is as follows:

[0070] (1) The instantaneous maximum temperature difference T in the battery pack during the entire discharge process F Determined as an inconsistency assessment indicator; T max_i =max{T 1_i ,T 2_i ...T n_i},i=1,2,3...k T min_i =min{T 1_i ,T 2_i ...T n_i},i=1,2,3...k ΔT max =max{T max_i -T min_i},i=1,2,3...k

[0071] Where, T n-i is the temperature value of the nth temperature sensor at time i, and the total time is k; T max-i and T min-i represents the maximum and minimum temperatures within the group at time i, ΔT max It is the instantaneous maximum temperature difference within the battery pack during the entire discharge process;

[0072] (2) The inconsistency evaluation index ΔT max The threshold T h The regulation is 5℃;

[0073] (3) Calculate the inconsistency score G of the pre-assembled battery pack, with a score range of [0-100];

[0074] The qualification condition for pre-assembled battery packs is that the inconsistency score G≤50. A qualified battery pack indicates that the pre-assembled single cells have good matching. Among them, G≤25 indicates that the battery pack has excellent consistency and can be used in high-demand equipment.

[0075] The temperature distribution and maximum temperature difference within the pre-assembled battery pack are as follows Figure 4 The temperature inconsistency of the three pre-assembled battery packs was calculated based on the temperature distribution within the packs. The calculation results are shown in Table 1.

[0076] Table 1 Inconsistency scores of pre-assembled parallel battery packs: Pre-assembled battery packs Group 1 Group 2 Group 3 Inconsistency score G 52.60 16.00 40.00

[0077] The inconsistency scores of Group 2 and Group 3 meet the requirements for parallel group use of batteries; however, the inconsistency score of Group 1 does not meet the conditions for parallel group use, and the next step of locating and replacing abnormal batteries is required.

[0078] Step 5: Based on the temperature data of each battery in the group obtained in the secondary test, use the Z-Score method to calculate the degree to which the average temperature of each battery deviates from the average temperature of other batteries. Calculate the Z-Score value of the branch battery based on the temperature data of each battery in the group obtained in the secondary test. The maximum value of |Z-Score| is the corresponding abnormal battery in the pre-assembled battery pack. The core formula is as follows:

[0079] Where: The average temperature of the jth battery; u group ,σ group are the mean and standard deviation of temperature within the group, respectively.

[0080] The calculated Z-Score values are shown in Table 2, where the battery with the maximum value of |Z-Score| (#3) is the abnormal battery in the group.

[0081] Table 2 Z-Score values of batteries in group 1 Battery label #3 #11 #13 Z-Score -0.0898 0.0196 0.0702

[0082] Replace the abnormal battery with the battery in the same cluster (#2) selected in step 3 (the new group is recorded as group 4), and continue the inconsistency detection in step 4. The temperature distribution and maximum temperature difference within group 4 are as follows Figure 4As shown in (d), the temperature inconsistency score calculation G (group 4) = 41.54, and its inconsistency score calculation G is greater than 50, which meets the secondary inspection requirements.

[0083] It should be noted that the three parallel battery packs are pre-assembled from batteries extracted from different clusters obtained from the first-level screening. Although group 1 showed unsatisfactory temperature inconsistency in the second-level inspection and needed to further locate and replace abnormal batteries, the necessity of the first-level screening cannot be ignored. The first-level screening quickly eliminated noise batteries (such as #5, #8, etc.) and divided the remaining batteries into clusters with similar parameters ( Figure 3 This process significantly narrows the candidate battery pool, providing a highly compatible candidate pool for secondary testing and avoiding the waste of resources required for full-scale testing. Furthermore, both Groups 2 and 3 passed secondary testing, demonstrating that primary screening can effectively predict the dynamic performance of battery packs.

[0084] While primary screening can increase the probability of group formation, it cannot completely replace actual operating condition verification. Secondary testing directly monitors temperature inconsistencies within the group during constant-current discharge, compensating for the fact that static parameters cannot reflect dynamic thermal behavior. Furthermore, secondary testing provides a basis for locating and replacing anomalies, forming a closed-loop process of "screening → verification → optimization."

[0085] In general, primary screening is a necessary prerequisite for secondary testing. The two together build a full-chain technical system of "static sorting → dynamic verification → precise optimization".

[0086] It should be noted that although the present invention has been described with reference to the above embodiments, the present invention may also have other various embodiments. Without departing from the spirit and scope of the present invention, it is obvious that those skilled in the art may make various corresponding changes and modifications to the present invention, and such changes and modifications shall fall within the scope of protection of the appended claims and their equivalents.

Claims

1. A secondary screening method for retired batteries in parallel groups, characterized in that: The steps include: Step S1: Perform performance tests on a batch of batteries with different health states, extract a number of internal electrochemical characteristic data of each battery, and use the extracted high-dimensional feature values after data standardization as screening features; Step S2: Mapping the high-dimensional feature data extracted from each single battery into a two-dimensional space using the t-SNE algorithm to obtain the two-dimensional features of each single battery; Step S3: Based on the two-dimensional features of each single battery obtained in step S2, the DBSCAN method is used to perform first-level cluster sorting on the batch of batteries to obtain battery groups in the same cluster with similar first-level cluster sorting positions, and noisy batteries with discrete first-level cluster sorting positions are eliminated; Step S4: pre-assemble the same cluster batteries selected in step S3 using a parallel device, and test the inconsistency of these batteries after parallel grouping for secondary inspection. Batches of batteries that meet the inconsistency requirements can be parallel grouped and used as needed; Step S5: For pre-assembled battery packs that fail the secondary test, the Z-Score method is used to locate mismatched batteries, and the abnormal batteries are replaced with the same cluster batteries screened in step S3. The inconsistency test in step S4 is continued until a batch of batteries with qualified parallel matching are screened out.

2. A secondary screening method for retired batteries in parallel groups according to claim 1, characterized in that: In step S1, the internal electrochemical characteristic data of each battery are extracted, which are maximum available capacity, ohmic impedance, charge transfer impedance, diffusion impedance and average voltage.

3. A secondary screening method for retired batteries in parallel groups according to claim 2, characterized in that: The maximum available capacity of the battery is the capacity released when the battery is fully charged and discharged at a constant current of 1C to the cut-off voltage; The ohmic internal resistance, charge transfer internal resistance, and diffusion impedance are obtained by fitting the first-order RC model through electrochemical impedance spectroscopy (EIS). Furthermore, the EIS curve is tested when the battery is at 50% SOC and the scanning frequency is 0.005 Hz to 100 kHz. The average voltage is the average voltage of the battery during the plateau period corresponding to 20%-80% SOC during 1C constant current discharge.

4. A secondary screening method for retired batteries in parallel groups according to claim 1, characterized in that: In step S1, the data normalization method selected is Z-Score normalization, and its core formula is: Where X is the original data point; u is the mean of the data set; and σ is the standard deviation of the data set.

5. The secondary screening method for retired batteries in parallel groups according to claim 1, characterized in that: The core steps of using the t-SNE algorithm to map the high-dimensional data feature matrix to a two-dimensional space in step S2 are as follows: Step S2.1, calculating high-dimensional space similarity; For a high-dimensional data point x i and x j , calculate x by the formula j As x i The conditional probability of the neighbor, reflecting x i The similarity with surrounding points is: Where σ i Represents the perplexity, controlling the number of effective neighbors for each point, where x i 、x j Represent the high-dimensional data features of different batteries labeled i and j respectively; Step S2.2, constructing a low-dimensional spatial probability distribution; In low-dimensional space, use the t distribution with 1 degree of freedom to calculate the low-dimensional data point y i and y j Similarity, the formula is: Where y i 、y j 、y k 、y l The high-dimensional data corresponding to batteries with different labels are mapped to low-dimensional data points of low-dimensional data; Step S2.3, minimizing KL divergence based on gradient descent to preserve the nonlinear relationship of the data; Calculate the KL divergence between the high-dimensional probability distribution P and the low-dimensional probability distribution Q. By minimizing the KL divergence, the low-dimensional space can reproduce the local structure of the high-dimensional space as much as possible. The smaller the KL divergence, the smaller the difference between the high-dimensional and low-dimensional distributions. The formula is: Based on the gradient descent method, the low-dimensional coordinates are optimized by calculating the KL divergence of the low-dimensional coordinates y i The gradient of the data point in the low-dimensional space is iteratively updated, and the low-dimensional distribution is gradually optimized to make it close to the high-dimensional distribution. The formula is:

6. A secondary screening method for retired batteries in parallel groups according to claim 1, characterized in that: The perplexity parameter of the t-SNE algorithm described in step S2 is set to 2-5.

7. The secondary screening method for retired batteries in parallel groups according to claim 1, characterized in that: In step S3, the neighborhood radius of the DBSCAN algorithm is set to 0.05-0.08, and the batteries within the neighborhood radius are marked as part of the same cluster battery group, and the batteries outside the neighborhood radius are marked as noise batteries.

8. The secondary screening method for retired batteries in parallel groups according to claim 1, characterized in that: In step S4, the inconsistency test condition after pre-assembly of the battery is to discharge the fully charged parallel battery pack at a constant current of 1C to the cut-off voltage at 25°C. The temperature data of each branch battery during the discharge process is collected by a temperature sensor. The inconsistency of the battery pack is described by the difference in temperature distribution. The specific inconsistency evaluation method is as follows: Step S4.1: calculate the instantaneous maximum temperature difference T F Determined as the inconsistency assessment indicator, the formula is as follows: T max_i =max{T 1_i ,T 2_i ...T n_i },i=1,2,3...k T min_i =min{T 1_i ,T 2_i ...T n_i },i=1,2,3...k ΔT max =max{T max_i -T min_i },i=1,2,3...k Where, T n-i is the temperature value of the nth temperature sensor at time i, and the total time is k; T max-i and T min-i represents the maximum and minimum temperatures within the group at time i, ΔT max It is the instantaneous maximum temperature difference within the battery pack during the entire discharge process; Step S4.2: Set the inconsistency evaluation index ΔT max The threshold T h The regulation is 5℃; Step S4.3, calculate the inconsistency score G of the pre-assembled battery pack, with a score range of [0-100], and the formula is: The qualification condition for pre-assembled battery packs is that the inconsistency score G≤50. A qualified battery pack indicates that there is good matching between the pre-assembled single cells, among which G≤25 indicates that the battery pack has excellent consistency and can be used in high-requirement equipment.

9. The secondary screening method for retired batteries in parallel groups according to claim 1, characterized in that: In step S5, the abnormal battery is located by using the Z-Score method to calculate the degree to which the average temperature of each battery deviates from the average temperature of other batteries. The Z-Score value of the branch battery is calculated based on the temperature data of each battery in the group obtained by the secondary detection. The maximum value of |Z-Score| is the corresponding abnormal battery in the pre-assembled battery pack. The formula is: Where, is the average temperature of the jth battery; u group ,σ group are the mean and standard deviation of temperature within the group, respectively.

10. The secondary screening method for retired batteries in parallel groups according to claim 1, characterized in that: If the iterative optimization described in step S5 fails to meet the requirements after more than three times, the entire battery pack will be downgraded for use in low-requirement scenarios.

Images