Retired lithium battery clustering method for evaluating service life

Abstract

The invention relates to a decommissioned lithium battery clustering method for evaluating service life, which comprises the following steps of: firstly, dividing batteries into short-life batteries and secondary usable batteries through an accelerated aging cycle test; the short-life batteries are classified into a class A and a class B according to the service life; collecting the discharge data of the first three cycles of the short-life battery as a training set; inputting the training set into a convolutional neural network for training so as to judge whether the two batteries are from the same group or not; collecting test sets, and comparing the test sets with the trained convolutional neural networks one by one; based on all the convolutional neural networks, calculating an adjacent matrix, grouping the batteries, and gathering the batteries with similar residual lives into one group; and finally, evaluating the performance of the battery pack in a normalized life loss mode. According to the method, the remaining service life of the battery is used as a battery classification feature, the batteries with similar remaining service life are clustered into one class, the service life loss of battery classification is reduced, and the secondary service life utilization rate of the retired battery is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power batteries, and in particular to a retired lithium battery clustering method for evaluating service life. Background Art

[0002] With the rapid development of the new energy industry, the scale of battery retirement is gradually increasing. The secondary use of retired batteries not only directly extends the battery life, but also alleviates the energy crisis and reduces environmental pollution. The technical challenge of battery reuse stems from the inherent low consistency of retired batteries, especially when considering the life span. Therefore, in order to ensure that the new battery pack can work properly and have a long service life, it is necessary to screen and exclude short-life battery cells before reassembling the battery pack.

[0003] There are two main methods for regrouping retired power lithium batteries: the first is regrouping through ordinary screening, which is mainly based on the acquired key characteristic factors, such as battery capacity, internal resistance, voltage, etc., and then screens the retired batteries according to predefined thresholds; the second is through clustering algorithms, which mainly use machine learning algorithms to regroup retired batteries with similar characteristics, thereby reducing the inconsistency in the battery pack. This method is suitable for occasions with a large number of retired batteries, and can achieve rapid classification and save costs. However, the classification algorithms that consider battery life are very limited, and the classification algorithms that use battery life as a feature are very limited. Therefore, the reorganization of the service life of retired batteries remains an unsolved problem in the field of power battery technology. Summary of the invention

[0004] The object of the present invention is to provide a battery classification method capable of reducing life loss, thereby improving the secondary life utilization rate of retired batteries.

[0005] To achieve the above object, the present invention proposes a retired lithium battery clustering method for evaluating service life, the method comprising the following steps:

[0006] S1: Classify the batteries into short-life batteries and reusable batteries through accelerated aging cycle test; and classify the short-life batteries into Class A and Class B according to their lifespan;

[0007] S2: collecting discharge data of the first three cycles of the short-life battery as a training set;

[0008] S3: inputting the absolute difference in voltage between each two battery discharge data in the training set into a convolutional neural network for training to determine whether the two batteries are from the same group;

[0009] S4: Collect a test set and compare the test set with the trained convolutional neural network one by one;

[0010] S5: Based on all convolutional neural networks, the adjacency matrix is ​​calculated to group the batteries, and batteries with similar remaining life are grouped together;

[0011] S6: Evaluate the performance of the battery pack by normalizing the life loss method.

[0012] Furthermore, the battery clustering method includes two parts: offline training and online testing. First, a batch of sample batteries are selected to perform accelerated aging cycle tests at the same temperature. Through the aging test data, a capacity decay curve is obtained. The capacity decay curve is observed, and the battery that first reaches the cutoff capacity under the same cycle is classified as a short-life battery. Then, the short-life battery is further divided into Class A and Class B according to the length of life. At the same time, other batteries that show better life are classified as reusable batteries. Then, the following offline training is performed: the discharge data of the 1st to 3rd cycles of the short-life battery (Class A and Class B) is used as a training set, and the absolute difference in voltage in the discharge data of each two batteries is selected as the input of the convolutional neural network, and the convolutional neural network is trained, wherein, by training n convolutional neural networks, it is determined whether the two batteries are from the same group. Then, the following online test is performed: the discharge data of the first three cycles of the unclassified batteries are used as the test set and compared one by one with the trained convolutional neural network; for the secondary reusable batteries, the input is the discharge data of the first three cycles after reaching the cutoff capacity; finally, the adjacency matrix is ​​calculated based on n convolutional neural networks so that batteries with similar remaining life can be grouped together.

[0013] Furthermore, step S1 is specifically as follows: subjecting a certain number of sample batteries to an accelerated aging cycle test at the same temperature; under the same cycle, the batteries that first reach the cutoff capacity are classified as short-life batteries, and those that do not reach the cutoff capacity are classified as secondary batteries; the short-life batteries are further divided into Class A and Class B according to their lifespan.

[0014] Furthermore, step S2 is specifically as follows: collecting discharge data of the first three cycles of the short-life battery, and taking the absolute voltage difference between every two batteries in the discharge data as a training set; the training set is collected in the following manner: 100 data points are uniformly selected in each cycle to form a 100×3 matrix; the first column of the 100×3 matrix is ​​the absolute voltage difference of the first cycle of the two batteries, the second column is the absolute voltage difference of the second cycle of the two batteries, and the third column is the absolute voltage difference of the third cycle of the two batteries; 100 data points represent 100 rows of the matrix.

[0015] Furthermore, in step S3, the training target is whether the two batteries are from the same group; the judgment result is represented by binary value, when the two battery groups are the same, it is true, which is [1; 0]; when the two battery groups are different, it is false, which is [0; 1].

[0016] Furthermore, in step S3, in order to minimize the errors caused by random network initialization and data segmentation in the training stage, the convolutional neural network is trained n times to obtain n classifiers, thereby improving the stability of the convolutional neural network model and reducing the errors caused by random network initialization and data segmentation; the 100×3 matrix is ​​input into the trained n classifiers, and the analogy ratio greater than 50% is the judgment result of the n classifiers; the final analogy ratio greater than 50% is divided into the corresponding class.

[0017] Furthermore, in step S4, the discharge data of the first three cycles of the unclassified battery or the discharge data of the first three cycles after the secondary battery reaches the cut-off capacity is used as a test set.

[0018] Furthermore, in step S5, the adjacency matrix is ​​calculated based on n convolutional neural networks to classify the batteries. Specifically, for each convolutional neural network, if battery #i and battery #j are predicted to be in the same group, then A i,j Just add 1, that is A i,j =A i,j +1; if predicted to be a different group, then A i,j Just subtract 1, that is, A i,j =A i,j -1; where, A={A i,j} is the adjacency matrix, A i,j represents the value corresponding to the row where battery #i is located and the column where battery #j is located in the adjacency matrix; A i,j ∈[-n, n], so that batteries that are expected to belong to the same category are clustered into a group and collected.

[0019] Furthermore, in step S6, the expression formula of the normalized life loss is:

[0020]

[0021] Among them, Loss is the normalized life loss of the battery pack, L j is the remaining service life of battery j, and BNo is the total number of batteries in the battery pack.

[0022] Compared with the prior art, the advantages of the present invention are:

[0023] 1. Compared with the traditional method based on capacity-internal resistance classification, the present invention innovatively proposes to use the remaining service life of the battery as a battery classification feature. The network used for battery training can also be used to classify retired batteries. Only by using the data of the first three charge and discharge cycles of the battery, the batteries can be classified into different groups, and batteries with similar remaining life can be clustered into one category, thereby reducing the life loss of battery classification and effectively improving the secondary life utilization rate of retired batteries.

[0024] 2. This invention highlights the promise of using supervised learning for battery classification tasks, providing new opportunities for the reliable use of retired batteries in second-life applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A flowchart of a retired lithium battery clustering method for evaluating life span according to an embodiment of the present invention;

[0026] Figure 2a A schematic diagram of a battery aging data set result of an accelerated aging cycle test in a method according to an embodiment of the present invention;

[0027] Figure 2b A schematic diagram of the life classification of short-life batteries of category A and category B in the method of an embodiment of the present invention;

[0028] Figure 2c A schematic diagram of test set data revised in the method of the embodiment of the present invention;

[0029] Figure 2d A schematic diagram of a data set of a training set and a test set in the method of an embodiment of the present invention;

[0030] Figure 3 This is an overall schematic diagram of the offline training and online testing parts in an embodiment of the present invention;

[0031] Figure 4 is an adjacency matrix of different batteries in the embodiment of the present invention;

[0032] Figure 5a A schematic diagram of battery classification results based on an adjacency matrix in a method according to an embodiment of the present invention;

[0033] Figure 5b A schematic diagram of battery classification results based on an adjacency matrix in a method according to an embodiment of the present invention;

[0034] Figure 5c This is a schematic diagram of a comparative example of the present invention based on a traditional capacity-internal resistance method;

[0035] Figure 5d Schematic diagram of battery classification results based on the traditional capacity-internal resistance method in the comparative example of the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be further described below.

[0037] This embodiment proposes a retired lithium battery clustering method for evaluating service life, such as Figure 1 As shown, the method includes steps S1-S6, wherein steps S2 and S3 are training parts, and steps S4 and S5 are testing parts. The specific method is as follows:

[0038] S1: Select a batch of new sample batteries and conduct accelerated aging cycle tests at the same temperature. The capacity decay curve is obtained through the accelerated aging cycle test. The capacity decay curve is observed. Under the same cycle, the batteries that reach the cutoff capacity first are classified as short-life batteries. These short-life batteries are further divided into Class A and Class B according to their lifespan. At the same time, other batteries that have not reached the cutoff capacity show better lifespan and are classified as secondary usable batteries.

[0039] In this embodiment, 38 18650 lithium batteries were selected for accelerated aging test, wherein the battery charging adopted constant current constant voltage curve, and the discharging adopted constant current curve, the constant current current was 7.5A, and the discharge cut-off voltage was 2.75V; the cut-off voltage and current in the charging stage were 4.20V and 0.13A respectively. After the accelerated aging test, the following results were obtained: Figure 2a The battery aging data set shown in FIG. 1 shows that, by observing the aging data set, it can be found that among the 38 batteries in this embodiment, 22 batteries reach the cut-off capacity of 2.15 Ah before 110 cycles, indicating that the life of these 22 batteries is relatively short and they will not be used for secondary use, and are classified as short-life batteries; while the other 16 batteries show better life behavior and are therefore selected as batteries that can be used for secondary use; in addition, as Figure 2b As shown, the 22 short-life batteries are further divided into Class A and Class B according to their lifespan.

[0040] S2: Collect the discharge data of the 1st to 3rd cycles of the above short-life batteries (type A and type B), such as Figure 2d The blue curve shown is the discharge data of the first cycle of all batteries in the training set; Figure 3 As shown, N is taken as 100, and 100 data points are uniformly selected in each cycle of the short-life battery to form a 100×3 matrix; the first column of the 100×3 matrix is ​​the absolute voltage difference of the first cycle of the two batteries, the second column is the absolute voltage difference of the second cycle of the two batteries, and the third column is the absolute voltage difference of the third cycle of the two batteries; the 100 data points represent 100 rows of the matrix.

[0041] S3: In this embodiment, if Figure 3As shown, two batteries are randomly selected from the two types of batteries, A and B, and the absolute difference in voltage in their discharge data is sent to the convolutional neural network for training. The training goal is whether the two batteries are from the same group. In order to minimize the errors caused by random network initialization and data segmentation in the training phase, n convolutional neural networks are trained. Therefore, the adjacency matrix will be calculated based on all n networks in the test part. In addition, the judgment result is represented by binary value: when the two battery groups are the same, it is true, which is [1; 0]; when the two battery groups are different, it is false, which is [0; 1].

[0042] S4: Collect a test set and compare the test set with the trained convolutional neural network one by one; the test set is the discharge data of the first three cycles of the unclassified battery; for the reusable battery, the discharge data of the first three cycles after it reaches the cut-off capacity is used as the training set; in this embodiment, the 111th to 113th cycles of the reusable battery are used as the discharge data, such as Figure 2d The red curve shown is a schematic diagram of the discharge data collected at the 111th cycle for all batteries in the test set.

[0043] S5: Based on all n convolutional neural networks, the adjacency matrix is ​​calculated to group the batteries, and batteries with similar remaining life are grouped together; in this embodiment, Figure 4 As shown, n is equal to 25, that is, 25 convolutional neural networks are used to group the batteries. For each convolutional neural network, if battery #i and battery #j are predicted to be in the same group, then A i,j Just add 1, that is A i,j =A i,j +1; if predicted to be a different group, then A i,j Just subtract 1, that is, A i,j =A i,j -1; where, A={A i,j} is the adjacency matrix; A i,j represents the value corresponding to the row where battery #i is located and the column where battery #j is located in the adjacency matrix; A i,j The initial value of A is 0. i,j ∈[-n, n], so that batteries that are expected to belong to the same category are grouped together and collected, such as Figure 4 Schematic diagrams of different adjacency matrices shown. For example, A 1,2 =7 means that at a certain matrix position ( Figure 4 The value of (row 1, column 11) is 7, which means that 16 networks (the majority of the 25 networks) believe that the two batteries (No. 01 and No. 02) can be classified into the same group; while 9 networks (the minority of the 25 networks) believe that the secondary lifespans of the two batteries are different.

[0044] In this embodiment, the classification results of the batteries are as follows: Figure 5a and 5b As shown, Figure 5a and Figure 5b It is a schematic diagram of the battery classification result according to the adjacency matrix in the method of the embodiment of the present invention. Figure 5a In the figure, the blue curve is the battery selected to be classified together; the red curve is the battery with a significant difference in initial capacity compared to the majority and can be easily removed from the battery pack; the gray curve is the remaining batteries that have not been classified together. Figure 5a It can be seen that batteries #02, #09, #10, #11, #13, #14, and #15 are grouped together, and battery #02 has a significantly lower initial capacity and can be removed from the battery pack more easily;

[0045] Figure 5b In the figure, the red curve represents the batteries that are selected and classified together. It can be seen that batteries #06, #07, #12, and #16 are grouped together. Battery #16 has a longer life, so grouping will not have a negative impact on the remaining three batteries.

[0046] At the same time, the traditional capacity-internal resistance classification method is used as a comparative example. Figure 5c As shown, the green circles represent the three batteries that are selected to be classified together based on capacity and internal resistance, and the black circles represent the remaining batteries that are not classified together.

[0047] The classification results based on the capacity-internal resistance classification method are as follows Figure 5d As shown, the green curve is the batteries that are selected and classified together, that is, batteries #04, #08, and #16 are grouped together; the gray curve is the remaining batteries that have not been classified together.

[0048] S6: In order to highlight the advantages of the method of the embodiment of the present invention over the traditional capacity-internal resistance method in the comparative example, the normalized life loss method is used to evaluate the performance of the battery pack in the embodiment method and the comparative example method, wherein the expression formula of the normalized life loss is:

[0049]

[0050] Among them, Loss is the normalized life loss of the battery pack, L j is the remaining service life of battery j, and BNo is the total number of batteries in the battery pack.

[0051] The evaluation results are shown in Table 1:

[0052] Table 1

[0053]

[0054] It can be directly seen from the above results that even if the initial capacity and resistance are very similar, the life of retired batteries may vary significantly, indicating that the capacity-internal resistance method may lead to incorrect battery grouping results. Specifically, when the battery #04, #08, and #16 are grouped together using the capacity-internal resistance method, their standardized life loss is equivalent to the loss of the original battery group #01 to #16 before classification. The method of the present invention can reduce at least 20% of the loss, and in the best case, it can even reduce 46% of the loss, which significantly demonstrates the effectiveness of the battery clustering method of this embodiment.

[0055] The above is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any technician in the relevant technical field, without departing from the scope of the technical solution of the present invention, makes any form of equivalent replacement or modification to the technical solution and technical content disclosed in the present invention, which does not depart from the content of the technical solution of the present invention and still falls within the protection scope of the present invention.

Claims

1. A method for clustering retired lithium batteries for evaluating service life, characterized in that: The following steps are involved: S1: Through accelerated aging cycle testing, the batteries are divided into short-life batteries and secondary-usable batteries; Classifying the short-life batteries into Class A and Class B according to their lifespan; S2: collecting the discharge data of the first three cycles of the short-life battery as a training set; S3: inputting the absolute difference in voltage between each two battery discharge data in the training set into a convolutional neural network for training to determine whether the two batteries are from the same group; S4: Collect a test set and compare the test set with the trained convolutional neural network one by one; S5: Based on the convolutional neural network, an adjacency matrix is ​​calculated to group the batteries, and batteries with similar remaining life are grouped together; S6: Evaluate the performance of the battery pack by normalizing the life loss method.

2. The retired lithium battery clustering method for evaluating service life according to claim 1, characterized in that: Step S1 is specifically as follows: subjecting a certain number of sample batteries to an accelerated aging cycle test at the same temperature; under the same cycle, the batteries that first reach the cutoff capacity are classified as short-life batteries, and those that do not reach the cutoff capacity are classified as secondary batteries; the short-life batteries are further divided into two categories, A and B, according to their lifespan.

3. The retired lithium battery clustering method for evaluating service life according to claim 1, characterized in that: Step S2 specifically includes collecting discharge data of the first three cycles of the short-life battery, and taking the absolute voltage difference between every two batteries in the discharge data as a training set; the training set is collected in the following manner: 100 data points are uniformly selected in each cycle to form a 100×3 matrix; the first column of the 100×3 matrix is ​​the absolute voltage difference of the first cycle of the two batteries, the second column is the absolute voltage difference of the second cycle of the two batteries, and the third column is the absolute voltage difference of the third cycle of the two batteries; 100 data points represent 100 rows of the matrix.

4. The retired lithium battery clustering method for evaluating service life according to claim 3 is characterized in that: In step S3, the judgment result is represented by binary value. When the two battery groups are the same, it is true, which is [1; 0]; when the two battery groups are different, it is false, which is [0; 1].

5. The retired lithium battery clustering method for evaluating service life according to claim 3, characterized in that: In step S3, the convolutional neural network is trained n times to obtain n classifiers; the 100×3 matrix is ​​input into the trained n classifiers, and the analogy ratio greater than 50% is the judgment result of the n classifiers; the final analogy ratio greater than 50% is divided into the corresponding category.

6. The retired lithium battery clustering method for evaluating service life according to claim 1, characterized in that: In step S4, the discharge data of the first three cycles of the unclassified battery or the discharge data of the first three cycles after the secondary battery reaches the cut-off capacity is used as a test set.

7. The retired lithium battery clustering method for evaluating service life according to claim 1, characterized in that: In step S5, specifically: for each convolutional neural network, if battery i and battery j are predicted to be in the same group, then A i,j =A i,j +1; if predicted to be a different group, then A i,j =A i,j -1; where A i,j represents the value corresponding to the row where battery #i is located and the column where battery #j is located in the adjacency matrix; A = {A i,j } is the adjacency matrix; A i,j ∈[-n, n], so that batteries that are expected to belong to the same category are clustered into one group.

8. The method for clustering retired lithium batteries for evaluating service life according to claim 1, characterized in that: In step S6, the expression formula of the normalized life loss is: Among them, Loss is the normalized life loss of the battery pack, L j is the remaining service life of battery j, and BNo is the total number of batteries in the battery pack.

Images