Research on Sorting Method of Lithium Ion Power Battery for Electric Vehicle
Abstract: In order to improve the inconsistency of the battery pack of electric vehicles and improve the available power and capacity utilization of the battery pack, based on the analysis of the inconsistency mechanism of the battery, the traditional sorting, main factor sorting and total factor are respectively performed on the battery. The calculation results of the monomer test show that the total factor sorting method is optimal. Then the fuzzy C-means clustering algorithm is used to sort the dynamic characteristics of the battery based on the total factor sorting results. The test results show that: The sorting method can effectively improve the inconsistency of the battery pack.
Keywords: vehicle engineering; lithium ion power battery; inconsistency; multi-parameter sorting; dynamic characteristic sorting
CLC number: U463.63 Document ID: A
Article ID: 1674-2974(2016)10-0023-09
Abstract:In order to improve the consistency for lithium-ion power battery of electric vehicle, to raise the available power and the Utilization ratio of capacity for the battery pack, this paper put forward a battery sorting method based on the inconsistency mechan Ism analysis of battery. Firstly, the traditional sorting method, the main factor sorting method and the total factor sorting method for battery were completed respectively, and the comparative results show that the total factor sorting method is optimal. Secondly, the dynamic characteristics sorting was Finally, the sorting effect was verified through experiments and the result shows that this sorting method can improve the inconsistency for battery pack effectiveness, and has a certain practical significance. Br\u003e Key words:vehicle engineering; lithium-ion power battery; inconsistence; multi-parameter sorting; dynamic characteristics sorting
In order to meet the requirements of driving motor supply voltage and vehicle driving range, the battery pack needs to pass hundreds of batteries The batterys are connected in series or in parallel. The inconsistency between the battery batterys causes the battery barrel to have a "cask effect" during use. The existence of this phenomenon will reduce the charging and discharging efficiency of the battery pack and reduce the driving range of the electric vehicle.
At present, the technical measures for solving the battery inconsistency mainly include battery sorting, battery balancing and battery thermal management. The battery sorting method is studied. The battery sorting method mainly includes single parameter sorting method, multi-parameter sorting method and dynamic characteristic sorting method, and the method of combining multi-parameter sorting method and dynamic characteristic sorting method is the current research. Trends.
Multi-parameter sorting method uses multiple characteristic parameters to sort the battery. Liu Qianjie et al  through 4 charge and discharge cycles, with a capacity difference of 1%, voltage of 10 mV, internal resistance of 2 mΩ and charging The constant current ratio is less than 1% for classifying the battery, and the sorting effect is better; Wu Shengxian  has a capacity tolerance of ±1% of the rated capacity, an open circuit voltage tolerance of ±10 mV, and an internal resistance tolerance of ±0.5 mΩ. The self-discharge tolerance is ±5 mV, and the sorting effect is also good; Jonghoon Kim et al. [3-4] proposed a capacity-based and internal resistance for the case where the voltage equalization technique may cause the voltage to be uniform and the state of charge is inconsistent. Come The method of voltage and SOC consistency of high lithium-ion power battery packs. However, the above multi-parameter sorting method does not optimize the sorting variables, and the correlation between sorting variables will affect the sorting results.
The characteristic sorting sorts the battery according to the charge-discharge curve. Shan Yi  obtains the difference between the batteries by using the hierarchical clustering method for the charge-discharge curve, and the test shows that the sorting effect of the method is better; Wen Tao  proposed a battery sorting method based on feature vector, but the standard voltage feature vector is more difficult to determine, which increases the difficulty of the sorting process. Yuan Fengyun  built the battery equivalent with SOC as the link. The circuit model sorts the battery according to the similarity between the charge and discharge curves obtained by the model simulation; Raspa et al.  sorts the battery according to the SOV change of the battery by self-organizing graph. However, the above dynamic characteristics The sorting method does not consider other parameters of the battery, and the implementation is more complicated.
This paper proposes a sorting method combining multi-parameter sorting and dynamic characteristic sorting. Multi-parameter sorting Based on the statistical software SPSS, the factor analysis module is used to optimize the sorting variables. The system clustering module is used to perform main factor sorting and total factor sorting on the battery. The dynamic characteristic sorting is based on the discharge curve of the battery. MATLAB programming to achieve fuzzy C-means clustering algorithm to achieve battery sorting.
1 Inconsistency analysis
Battery inconsistency refers to the same batch, the same specification, the same type of battery, in the voltage Differences in characteristic parameters such as capacity, internal resistance, and self-discharge rate.
The inconsistency of lithium-ion power batteries is mainly generated during production, use, and storage. Battery inconsistency caused during production It is accumulated and expanded during use and storage. For example, two batteries with different capacities are charged and discharged in series, the current passing through is the same, the charge and discharge capacity is the same in the same time, and the battery with the small capacity reaches its limit capacity. At this time, the battery with a large capacity may be in an unfilled or uncharged state, which will result in wasted energy. Line charging and discharging will inevitably make the battery with small capacity always in deep charge and deep discharge state, while the battery with large capacity is always in shallow charge and shallow state, so that the performance of the battery with small capacity will become worse and worse, and the battery with large capacity The inconsistency will be further aggravated.
So it is especially important to sort the battery before using it in a reasonable way.
2 Multi-parameter sorting
This article first based on the statistical software SPSS Multi-parameter sorting of batteries.
2.1 Acquisition of sorting variables
The 100-battery 180 Ah battery is used as a research object to explore effective battery sorting methods. The technical parameters are shown in Table 1.
Before the multi-parameter sorting, the sorting variable needs to be determined. The sorting variable can be selected based on the inconsistency expression form. In addition, the charging process of the battery includes constant current charging and constant voltage charging, and generally the constant current charging is performed first. Then, constant voltage charging is performed, the constant current process is a process of generating polarization, and the constant voltage process is a process of eliminating polarization, and the shorter the constant voltage process time, indicating that a constant current process is generated. The smaller the polarization, the better the battery performance. Finally, the average internal resistance, open circuit voltage, self-discharge rate, charging capacity, discharge capacity and the ratio of constant current charging time to total charging time are selected as the sorting variables of the battery. Among them, the average internal resistance is the average value of the internal resistance of the battery when 1/3 C is fully charged and 1/3 C is discharged, and is recorded as R; the open circuit voltage is the value of the battery at 1/3 C full charge 3 The voltage value after d is recorded as D; the self-discharge rate is the voltage drop of 7 d after the battery is fully charged at 1/3 C under 40 °C, and is recorded as E; the charge and discharge capacity are respectively taken as the battery at 1 /3 C standard charge and discharge capacity, recorded as Q1, Q2; constant current charging time as a percentage of total charging time The ratio of time is recorded as B.
In order to obtain the test data of the sorting variables, the NEWARE charging and discharging device is used to charge and discharge 100 batterys respectively. Among them, the charging and discharging test photos of the battery are shown in Fig. 1. Br\u003e 2.2 Optimization of Sorting Variables
To eliminate the phase between sorting variables The effect of the correlation on the sorting results, while reducing the sorting variables, simplifying the calculation, can factor the battery sorting variables .
Factor analysis is a statistical method, the most commonly used theoretical formula As shown in formula (1):
The statistical software SPSS uses the correlation coefficient matrix, the reflection image correlation matrix, the Bartlett sphericity test and the KMO test as the conditions for the judgment factor analysis. Among them, it is often used. Bartlett sphericity test and KMO test.
Bartlett spheroid test tests the correlation matrix to check whether the battery sorting variable is suitable for factor analysis. The original assumption of Bartlett spherometry is that the correlation matrix is ??a unit matrix, and only the assumption is rejected. Factor analysis is meaningful. To reject this hypothesis, the corresponding probability value Sig of Bartlett spheroid statistic is less than the given significance level. The KMO test is judged by the correlation coefficient between the battery sorting variables. Whether the selected variable is suitable for factor analysis. The larger the KMO value, the larger the correlation coefficient between the sorted variables, the more common they are. Usually, the KMO value reaches 0.7. Factor analysis can be used above.
Since the Bartlett sphericity test and the KMO test are related to the correlation matrix between the sorting variables, it is first necessary to obtain the correlation matrix between the sorting variables. The experimental data of the sorting variables are taken as input, and the correlation matrix of the six sorting variables is calculated as shown in Table 2.
It can be seen from Table 2 that there is a large correlation between the sorting variables of the battery. Therefore, it is necessary to perform Bartlett test and KMO test on the correlation matrix. The test results obtained are shown in Table 3. It can be seen that the Bartlett sphericity test statistic corresponding probability value Sig=0.000, less than the given significance level of 0.050 and KMO=0.896\u003e0.700 satisfies the factor analysis conditions, so the selected sorting variables can be optimized by factor analysis.
The principal component analysis method is selected to analyze the selected six sorting variables, and the results are shown in Table 4. The contribution rate of the principal component. The contribution rate of the principal component indicates the degree to which the factors obtained by principal component analysis can explain the original sorting variable.
Usually the factor with the eigenvalue greater than 1 is taken. As a new variable representing the original variable . In Table 4, there are three factors whose eigenvalues ??exceed 1, so this factor is taken as the new sorting variable, ie the main factors F1, F2 and F3. Visible, sorting variables After factor analysis, 3 principal factors can be extracted to express 86.834% of the content.
The obtained factor matrix is ??shown in Table 5.
The factor matrix in Table 5 is the value of each raw sorting variable in each factor. The factor load above, for example, the discharge capacity = 0.970 × F1 - 0.049 × F2 + 0.094 × F3. From the factor load in the table, the first factor mainly expresses the discharge capacity, open circuit voltage, charge capacity, and constant current charge. The ratio of time to total charge time is the four sorting variables. The second factor mainly expresses the two sorting variables, the average internal resistance and the self-discharge rate, while the third factor comprehensively expresses each sorting. Variables.
The obtained factor score coefficient matrix is ??shown in Table 6.
The data in each column of Table 6 is the coefficient represented by the original sorting variables. For example: main factor F1=0.030 × average internal resistance - 0.320 × open circuit voltage + 0.012 × self-discharge rate + 0.252 × Capacitance +0.320 × discharge capacity +0.265 × constant current charging time as a percentage of total charging time.
In summary, the first six sorting variables are converted into three variables (ie, three principal factors) after factor analysis. It can express most of the information of the original sorting variable, so these three main factors are selected as the new sorting variables. In addition, according to the degree to which each main factor can express the original sorting variable, it can be integrated into a total factor. The weighting factor of this total factor is determined according to Table 4, that is, the total factor F=49.434%×F1+20.667%×F2+16.733%×F3.
2.3 Sorting results of batteries
Cluster analysis is direct comparison The nature of things, classify things with similar properties into one class, and things with different natures fall into different classes of technology.
The distance between samples in cluster analysis and between samples and classes, classes and classes The calculation method of distance is very important. In view of the wide application of the square Euclidean distance metric, this distance is chosen as the measure of the distance between samples. The expression of this distance is shown in equation (2). Good classification in application The application is widely used, so this method is selected to cluster the battery.
The following is a systematic clustering method for main factor sorting and total factor sorting of battery samples, and compared with the traditional sorting method. In order to facilitate the verification of the sorting effect, all the batteries are divided into 4 categories when sorting according to the following method.
1) Traditional sorting method
Traditional sorting method directly follows the traditional practice of battery manufacturers The discharge capacity, internal resistance and open circuit voltage are used as sorting variables, that is, the battery batterys are selected first according to the discharge capacity, the internal resistance and the open circuit voltage, and then the difference between the capacity group, the internal resistance group, and the voltage group difference. Grouping, the sorting results obtained according to this method are shown in Table 7.
2) Main factor sorting method
The main factor sorting method is the three main factors generated by factor analysis in Section 3.2. F1, F2, F3 are used as sorting variables, and the battery clustering is performed by the system clustering method in SPSS software. Among them, the distance between samples is measured by the square Euclidean distance, and the clustering method is selected by the squared method of dispersion. The results of the sorting method are shown in Table 8.
3) Total factor sorting
The total factor sorting method is to use the total factor F as a sorting variable, and cluster it by systematic clustering in SPSS software. Among them, the measurement method of distance between samples still uses square Euclidean distance, and the clustering method still uses the method of squared deviation. The sorting results are shown in Table 9.
2.4 Comparison of multi-parameter sorting effect
Since the initial data of each battery sorting variable is known, the sorted batteries can be evaluated based on these initial data. As shown in Table 10, the average discharge capacity and average self-discharge rate of each type of battery are used. To evaluate the performance of such batteries.
It can be seen from Table 10 that the conventional sorting method only uses internal resistance, open circuit voltage and discharge capacity as sorting variables, and does not consider the influence of self-discharge rate and other factors, so each The average discharge capacity of the battery is relatively high, but the average self-discharge rate is also high, and the distribution may be uneven. The main factor sorting method considers more factors, but it does not follow each main factor. For the original The degree of interpretation of the selected variables is weighted. The average discharge capacity and the average self-discharge rate of each type of battery are centered, while the factors for total factor sorting are more comprehensive, and the original sorting variables are selected according to each main factor. The degree of interpretation is weighted, so the average discharge capacity of each type of battery sorted by it is higher, the average self-discharge rate is lower, and the difference between class and class is larger, and the sorting effect is the best. Br\u003e 3 Dynamic Characteristics Sorting
The multi-parameter sorting method is static sorting. Although it can reflect some characteristics of the power battery, it is mainly external characteristics, and it can not reflect the change trend of battery characteristics during charging and discharging. The dynamic characteristic sorting method is based on the discharge curve. Considering the different internal structure of the battery during charging and discharging, combined with multi-parameter sorting, it is possible to select a battery with better consistency, thereby improving the performance of the battery pack. Br\u003e 3.1 Discrete fitting of battery charge and discharge curves
From the comparison of multi-parameter sorting results, the total factor sorting method is the best among the three multi-parameter sorting methods, but The Class III batteries that are sorted out are relatively inferior. The Class IV batteries have only one section, so the dynamic characteristics can only be based on the Class I (29) and Class II (53) batteries. Sorting.
Selecting p sampling points on the battery charge and discharge curve, the charge and discharge curve of the battery can be converted into a one-dimensional feature vector. For a group of n batteries to be classified, a cluster charge and discharge curve can be used. Converted into an original data array of n × p dimensions.
The difference in performance between batteries will result in different charging and discharging times of the battery. In order to make the length of the feature vector representing each battery the same, it is convenient to calculate the distance between the batteries. It is necessary to fit the charge and discharge data of each battery. Here, only the rest and discharge sections are performed, and 100 data points and 500 discharge data points representing each battery are obtained, and a total of 600 data points are used to represent one section. Battery.
3.2 Dynamic Characteristics of Battery Sorting
The clustering method is needed to sort the dynamic characteristics of the battery. In this paper, the fuzzy C-means clustering algorithm is used to divide the battery according to the charge and discharge curve of the battery. Fuzzy fuzzy C-means clustering algorithm is an objective function-based clustering algorithm proposed by Dr. Jim Bezdek in 1973 . The membership function, similarity function and objective function are simple. Introduction.
1) Membership function
The membership function is used to indicate the extent to which a battery belongs to a certain type of battery. It is represented by uA(X). If its value is 1, it means that the battery is completely Belongs to a certain type of battery. According to the normalization rule, the sum of the membership degrees of a data set is equal to 1, that is, if there are n samples, c cluster centers, then
2) similarity function
The squared Euclidean distance is chosen as the similarity function. The smaller the distance, the more similar the two samples are.
3) The objective function
These two necessary conditions make the fuzzy C-means clustering algorithm an iterative process. The process is as follows: First, it is necessary to initialize U and make it satisfy the sum to be 1; secondly, calculate c ci by formula (8); finally calculate the objective function according to formula (6), if the condition is satisfied, the algorithm stops, if Not satisfied, Then use the formula (9) to calculate the new U, and then calculate from the beginning.
According to the introduction of the fuzzy C-means clustering algorithm, the n-battery samples are clustered into the c-class based on the total factor sorting method. Verification, taking the final selected number of battery segments m=4 as an example, the algorithm flow is shown in Figure 2.
According to the flow shown in Figure 2, the algorithm is implemented by MATLAB programming, and the dynamic characteristic sorting is obtained. The result of the method sorting.
Taking the first type of battery (29 knots) as an example, since the number of battery sections is large, it is first divided into three categories, and then fuzzy C-means clustering is performed for each class to obtain The clustering result is shown in Fig. 3. The objective function J is iterated 13 times and reaches 1e-5, which satisfies the set condition and the iteration stops.
As shown in Fig. 3, the second category of the 29-battery battery contains only 2 batterys can be discarded. Classes 1 and 3 contain 13 and 14 batterys respectively, and they can continue to be clustered separately.
1) Clustering results of 13 batterys:\u003cbr \u003e When clustered into 3 categories, the clustering results as shown in Figure 4 are obtained.
When clustered into 4 categories, the cluster shown in Figure 5 is obtained. Class results.
It is known from Fig. 4 and Fig. 5 that the four batteries of 61, 78, 80, and 87 are always grouped together regardless of whether they are clustered into three or four types, so the consistency of the four batteries can be considered. Better.
2) Clustering results of 14-battery battery:
When clustered into 3 categories, the clustering result shown in Figure 6 is obtained.
When clustered into 4 categories, the graph is obtained. The clustering result shown in Fig. 7
It can be seen from Fig. 6 and Fig. 7 that the first type of battery (6 sections) when clustered into 4 types includes the first type of battery (8 sections) when clustered into 3 types. In order to select a 4-battery battery with higher consistency, the first type of battery when clustered into 4 types continues to be clustered.
For these 6-battery batteries, when the number of clusters is 3, the figure is as shown in the figure. The clustering result shown in Fig. 8. It can be seen from Fig. 4 and Fig. 8 that the number of batteries in each type is less than 4 knots, and the number of batteries in the final cluster is uniformly 4, so it is unreasonable to think that the number of clusters is 3. When the clustering is 2, the clustering results as shown in Fig. 9 are obtained. It can be seen that the 10 batteries of 10, 15, 28, 90 are grouped together. Therefore, among the 14 batteries, 10 The consistency of the four batteries of 15,28,90 is better.
From the above clustering process, in the 29-battery battery, the two groups of batterys 61, 78, 80, 87 and 10, 15, 28, 90 have good consistency.
Similarly, by total factor sorting method The dynamic characteristics of the sorted Class II batteries (53 knots) are similar to those of the Class I batteries (29), and are not described here. Summarize the sorting results of these two types of batteries as shown in Figure 10 and 11 shows. In summary, on the basis of total factor sorting, 7 groups of batteries with good consistency were selected by dynamic characteristic sorting, which are labeled as 61, 78, 80, 87; , 15, 28, 90; 12, 13, 30, 36; 84, 85, 86, 96; 60, 73, 93, 94; 5, 20, 68, 69; 2, 3, 42, 46 batteries. Br\u003e 4 test verification
In order to verify the effectiveness of the combination of multi-parameter sorting and dynamic characteristic sorting, the 29- and 53-battery batteries were separately sorted according to the total factor sorting method. The number of selected battery segments is 4, and the number of groups is 7 as the sorting target. The numbers of the 7 sets of batteries are 6,61,80,82; 1,72,78,97; 2,8,33,85; 12,13 68,94;20,46,47,89;16,29,81,92;30,55,83,96.
The selected batteries are connected in series to perform a charge and discharge test to discharge voltage The difference is used as the evaluation index, and the test results are shown in Table 11.
It can be seen from Table 11 that the discharge voltage difference of each group of batteries sorted by the dynamic characteristic sorting method after the total factor sorting is 0.024 8 V The discharge voltage difference of each battery that has been sorted by the total factor sorting method is 0.042 7 V. The sorting conditions of the grouped power battery are usually set as follows: the range is ≤0.1 V. Visible, two points The selection method satisfies the requirements, but the method of dynamic characteristic sorting after total factor sorting is more effective.
1) The sorting variables of the battery are optimized by factor analysis, through system clustering The batteries were clustered. The experimental results show that the sorting effects of the three multi-parameter sorting methods are from the good to the bad, the total factor sorting method, the main factor sorting method, and the traditional sorting method.
2) Based on the results of the total factor sorting, the dynamic characteristics of the battery were sorted. The test results show that the results are known. The discharge voltage difference of each group of batteries sorted by the dynamic characteristic sorting method after sorting by the total factor is 0.024 8V, which proves that the combination of the total factor sorting method and the dynamic characteristic sorting method proposed in this paper is effective. Sex.
 Liu Qianjie, Liu Keda, Kong Xianghua. A Lithium Ion Battery Sorting Method: China, 201410154729.4[P].2014-07-30.
LIU Qian-jie , LIU Ke-da, KONG Xiang-hua. A sorting method for lithium-ion battery: China, 201410154729.4[P].2014-07-30.(In Chinese)
 Wu Shengxian. Lithium-ion battery Selection method: China, 201210405058.5[P]. 2014-05-07.
WU Sheng-xian. Sorting method for lithium-ion battery: China, 201210405058.5[P].2014-05-07. (In Chinese)\u003c Br\u003e  KIM J, SHIN J, JEOM C, et al. High accuracy state-of-charge estimation of Li-Ion battery pack based on screening process [C]//Applied Power Electronics Conference and Exposition, 2011 Twenty- Sixth Annual IEEE. IEEE, 2011: 1984-1991.
 KIM J, SHIN J CHUN C, et al. Stable configuration of a Li-ion series battery pack based on a screening process for improved voltage/SOC balancing[J]. Power Electronics, IEEE Transactions on, 2012, 27(1): 411-424. Br\u003e  Shan Yi. Lithium-ion battery consistency study [D]. Shanghai: Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences, 2008.
SHAN Yi. Research on uniformity of lithium-ion battery [D] .Shanghai: Institute of Microsystem and Information Technology, Chinese Academy of Sciences, 2008. (In Chinese)
 Wen Tao, Cai Hao. A Li-ion Battery Matching Method Based on Battery Feature Vector: China, 201110367124 .X[P]. 2012-03-28.
WEN Tao, CAI Hao. A lithium-ion batteries group method based on feature vector:China,201110367124.X[P]. 2012-03-28. (In Chinese)
 Yuan Fengyun. Study on the separation method of lithium ion battery based on charge and discharge characteristics [D]. Changchun: School of Transportation, Jilin University, 2014.
YUAN Feng-yun. Research on lithium-ion Batteries sor Ting method based on characteristics of charging and discharging [D]. Changchun: College of Transportation,Jilin Univ, 2014. (In Chinese)
 RASPA P, FRINCONI L, Adriano MANCINI, et al. Selection of lithium batterys For EV battery pack using selforganizing maps[J]. J Automotive Safety and Energy, 2011, 2(2): 157-164.
 Zhang Wenzhao, Yan Hao. Basic Course of SPSS Statistical Analysis [M]. Beijing: Higher Education Press, 2004.
ZHANG Wen-tong, YAN Hao. SPSS statistical analysis basic tutorial[M].Beijing:Higher Education Press,2004. (In Chinese)
KANNAN SR, RAMATHILAGAM S , CHUNG P C. Effective fuzzy c-means clustering algorithms for data clustering problems [J]. Expert Systems with Applications, 2012, 39 (7): 6292-6300.