A Grouping Method for Lithium-ion Batteries Based on Random Forest Algorithm

Abstract

This invention relates to a lithium-ion battery grouping method based on the random forest algorithm, belonging to the field of lithium-ion battery technology. The method uses discharge curves as features for grouping, and employs the random forest algorithm for classification during the grouping process, grouping batteries with similar characteristics together. Experimental results demonstrate that, compared to other state-of-the-art clustering solutions, the grouping method proposed in this invention achieves more accurate grouping.

Description

Technical Field

[0001] This invention belongs to the field of lithium-ion battery technology and relates to a grouping technology for determining the batteries of underwater vehicles during the installation process, specifically a lithium-ion battery grouping method based on the random forest algorithm. Background Technology

[0002] Batteries are one of the most critical enabling technologies for achieving clean, efficient, and sustainable energy development in the transportation and power sectors. The small capacity and low voltage of a single battery cell cannot meet the demands of electric vehicles and other battery storage applications. Therefore, a group of battery cells are combined to form battery modules or battery packs, providing sufficient power and energy storage capacity for certain applications. However, due to limitations in manufacturing processes, different batteries exhibit inconsistent characteristics. This inconsistency can lead to some batteries in a battery pack being overcharged or over-discharged, while others remain underutilized. This can severely impact the overall performance and lifespan of the battery pack. Furthermore, battery inconsistency also presents greater challenges for monitoring and managing the battery pack's condition.

[0003] Differences in raw materials and manufacturing processes are the primary cause of parameter inconsistencies. To address these inconsistencies, battery manufacturers should improve the consistency and quality of battery materials and the battery manufacturing process. However, improving or optimizing battery material consistency is typically an expensive process. In contrast, battery grouping is a more efficient way to handle inconsistencies because it groups batteries with similar characteristics within the same battery pack.

[0004] In actual battery production, the battery grouping process still relies on a traditional manual grouping method based on multiple measured voltages. This manual grouping method is labor-intensive and inexpensive. However, it is also prone to errors due to human negligence. In recent years, with the rapid development of automation technology, more and more automation technologies are being used to replace traditional manual methods to improve efficiency and reduce costs.

[0005] Generally, traditional clustering algorithms are unsuitable for this problem. For example, the k-means clustering algorithm is an iterative clustering analysis algorithm. Its steps are: first, divide the data into K groups; then, randomly select K objects as initial cluster centers; calculate the distance between each object and each seed cluster center; and assign each object to the nearest cluster center. The cluster centers and the objects assigned to them represent a cluster. Each time a sample is assigned, the cluster centers are recalculated based on the existing objects in the cluster. This process is repeated until a termination condition is met. The termination condition could be that no objects are reassigned to different clusters, no cluster centers change, or the sum of squared errors reaches a local minimum. While the k-means clustering algorithm is simple to implement and converges quickly, the value of k is difficult to determine, and it is susceptible to noise, edge points, outliers, etc. Battery data has nonlinear and noisy characteristics, therefore this method is not suitable for battery grouping. The Random Forest algorithm constructs a CART classification tree and iteratively obtains partitioning features by traversing all attribute dimensions of the data sample set to achieve grouping. Because the Random Forest algorithm consists of multiple CART trees, it has certain advantages in handling nonlinear problems and noise resistance problems. Summary of the Invention

[0006] The technical problem to be solved by this invention is:

[0007] To address the limitations of traditional battery grouping methods, this invention proposes a lithium-ion battery grouping method based on the random forest algorithm, and implements it on a host computer to achieve better battery grouping.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0009] A lithium-ion battery grouping method based on the random forest algorithm, characterized by the following steps:

[0010] S1: Conduct a charge-discharge test on the battery to obtain the discharge curve;

[0011] S2: Sample two voltage values ​​at different times in the discharge curve of each battery, namely the charging static voltage and the discharge termination voltage. The two sampled values ​​of each battery are its characteristic values.

[0012] S3: Calculate the Gini index required for the Random Forest algorithm.

[0013] The Gini index is calculated as follows:

[0014]

[0015] Where D represents the data sample set, its dimension is denoted as M*N, M represents the number of batteries to be classified, and N represents the feature values ​​sampled for each battery, namely the charging static voltage and the discharge termination voltage; M k This represents the number of batteries in the k-th group after grouping.

[0016] S4: Constructing a CART classification tree based on the Gini index;

[0017] S4.1: Obtain the Gini index of all current battery sample sets;

[0018] S4.2: Traverse all attribute dimensions of the data sample set and randomly select k feature sets of batteries as the splitting values;

[0019] S4.3: Calculate the Gini index of each battery after division and obtain the corresponding reduction in Gini index;

[0020] S4.4: Take the maximum reduction in Gini index among all batteries as the optimal partition and record the corresponding feature dimension and feature partition value;

[0021] S4.5: Set the threshold for reducing the Gini index. If the current reduction in the Gini index is less than the set threshold, or the number of splits reaches the set upper limit, then stop splitting and the CART classification tree is completed. Otherwise, according to the current best split, divide all data samples with feature values ​​less than the current best split value into the left tree and data samples with feature values ​​greater than the current best split value into the right tree, and save them to the node class of the current tree. Then jump to S4.1 to continue splitting the left and right subsets.

[0022] Step 5: After the CART classification tree in S4 is constructed, repeat step S4 m times to form m CART trees, thus forming a random forest. For each battery classification, these m decision trees will classify the battery, and the classification result with the highest consistency will be the final grouping result.

[0023] A further technical solution of the present invention: S1 specifically includes:

[0024] First, place the batteries on the testing equipment and charge each battery individually using a constant current until it reaches the charging termination voltage. Then, allow the batteries to rest for 24 hours. Next, discharge each battery individually using a constant current until a fixed discharge time is reached. Measure the voltage of the discharged battery at fixed intervals and upload the data to the host computer to obtain the discharge curve.

[0025] A further technical solution of the present invention: the constant current is 0.1C, where C refers to the charge / discharge rate.

[0026] A further technical solution of the present invention: the fixed time is 10 seconds.

[0027] A computer system is characterized by comprising: one or more processors, and a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method described above.

[0028] A computer-readable storage medium is characterized by storing computer-executable instructions, which, when executed, are used to implement the above-described method.

[0029] Beneficial effects:

[0030] This invention provides a lithium-ion battery grouping method based on the random forest algorithm. The method uses the discharge curve as a feature for grouping and employs the random forest algorithm for classification during the grouping process to group batteries with similar characteristics together.

[0031] For a group of batteries to be grouped, the goal is to maximize the number of successfully grouped battery packs by making full use of batteries with similar characteristics. In the method of this invention, the batteries in the battery pack have more similar characteristics than other batteries and meet the characteristic consistency criteria.

[0032] Experimental results demonstrate that, compared with other state-of-the-art clustering solutions, the grouping method proposed in this invention can achieve more accurate grouping. Attached Figure Description

[0033] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0034] Figure 1 Flowchart of the method of this invention;

[0035] Figure 2 Discharge curve;

[0036] Figure 3 CART classification tree construction flowchart;

[0037] Figure 4 Random Forest Construction Flowchart;

[0038] Figure 5 Figures showing the accuracy verification results of the method of this invention: (a) Support Vector Machine; (b) Random Forest Algorithm. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0040] The discharge curve of a battery can well reflect its capacity and characteristics. Therefore, this invention makes full use of this and uses the battery discharge curve as a feature to complete the grouping process.

[0041] This invention provides a lithium-ion battery grouping method based on the random forest algorithm, such as... Figure 1 As shown, it includes the following steps:

[0042] S1: First, place the batteries on the testing equipment and charge each battery individually using a constant current (0.1C, where C refers to the charge / discharge rate) until it reaches the charging termination voltage. Let the batteries rest for 24 hours. Then, discharge each battery individually again using a constant current (0.1C) until the fixed discharge time is reached. Measure the voltage of the discharged battery every 10 seconds and upload the data to the host computer to obtain the discharge curve, such as... Figure 2 As shown.

[0043] S2: After obtaining the discharge curve of each battery, sample two voltage values ​​at different times in the discharge curve of each battery, namely the charging static voltage and the discharge termination voltage. The two sampled values ​​of each battery are its characteristic values.

[0044] S3: Calculate the Gini index required for the Random Forest algorithm.

[0045] The Gini index is calculated as follows:

[0046]

[0047] Where D represents the data sample set, its dimension is denoted as M*N, M represents the number of batteries to be classified, and N represents the feature values ​​sampled for each battery, namely the charging static voltage and the discharge termination voltage. k This represents the number of batteries in the k-th group after grouping.

[0048] S4: Combining the Gini index calculation method, the process of constructing a CART classification tree based on the Gini index is as follows: Figure 3 As shown.

[0049] S4.1: Obtain the Gini index of all current battery sample sets;

[0050] S4.2: Traverse all attribute dimensions of the data sample set and randomly select k feature sets of batteries as the splitting values;

[0051] S4.3: Calculate the Gini index of each battery after division and obtain the corresponding reduction in Gini index.

[0052] S4.4: Take the maximum reduction in Gini index among all batteries as the optimal partition and record the corresponding feature dimension and feature partition value.

[0053] S4.5: Set a threshold for reducing the Gini index. If the current reduction in the Gini index is less than the set threshold, or the number of splits reaches the set upper limit, then splitting stops, and the CART classification tree is completed. Otherwise, according to the current best split, all data samples with feature values ​​less than the current best split value are divided into the left tree, and data samples with feature values ​​greater than the current best split value are divided into the right tree, and saved to the node class of the current tree. Then jump to S4.1 to continue splitting the left and right subsets.

[0054] S5: After the CART classification tree in S4 is constructed, repeat step S4 m times to form m CART trees, thus forming a random forest. For each battery classification, these m decision trees will classify the battery, and the classification result with the highest consistency will be the final grouping result.

[0055] To verify the superior performance of the algorithm proposed in this invention, the proposed method is compared with the support vector machine algorithm, such as... Figure 5 As shown in the figure, the small circles in the figure represent the number of batteries that failed to achieve effective classification. It can be seen from the figure that the random forest algorithm is significantly better than the support vector machine algorithm.

[0056] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the scope of the technology disclosed in the present invention, and such modifications or substitutions should all be covered within the scope of protection of the present invention.

Claims

1. A lithium-ion battery grouping method based on the random forest algorithm, characterized in that... The steps are as follows: S1: Conduct a charge-discharge test on the battery to obtain the discharge curve; S2: Sample two voltage values ​​at different times in the discharge curve of each battery, namely the charging static voltage and the discharge termination voltage. The two sampled values ​​of each battery are its characteristic values. S3: Calculate the Gini index required for the Random Forest algorithm. The Gini index is calculated as follows: Where D represents the data sample set, its dimension is denoted as M*N, M represents the number of batteries to be classified, and N represents the feature values ​​sampled for each battery, namely the charging static voltage and the discharge termination voltage; M k This represents the number of batteries in the k-th group after grouping. S4: Constructing a CART classification tree based on the Gini index; S4.1: Obtain the Gini index of all current battery sample sets; S4.2: Traverse all attribute dimensions of the data sample set and randomly select k feature sets of batteries as the splitting values; S4.3: Calculate the Gini index of each battery after division and obtain the corresponding reduction in Gini index; S4.4: Take the maximum reduction in Gini index among all batteries as the optimal partition and record the corresponding feature dimension and feature partition value; S4.5: Set the threshold for reducing the Gini index. If the current reduction in the Gini index is less than the set threshold, or the number of splits reaches the set upper limit, then stop splitting and the CART classification tree is completed. Otherwise, according to the current best split, divide all data samples with feature values ​​less than the current best split value into the left tree and data samples with feature values ​​greater than the current best split value into the right tree, and save them to the node class of the current tree. Then jump to S4.1 to continue splitting the left and right subsets. Step 5: After the CART classification tree in S4 is constructed, repeat step S4 m times to form m CART trees, thus forming a random forest. For each battery classification, these m decision trees will classify the battery, and the classification result with the highest consistency will be the final grouping result.

2. The lithium-ion battery grouping method based on random forest algorithm according to claim 1, characterized in that: S1 specifically refers to: First, place the batteries on the testing equipment and charge each battery individually using a constant current until it reaches the charging termination voltage. Let the batteries rest for 24 hours. Then, discharge each battery individually using a constant current until a fixed discharge time is reached. Measure the voltage of the discharged battery at fixed intervals and upload the data to the host computer to obtain the discharge curve.

3. The lithium-ion battery grouping method based on the random forest algorithm according to claim 2, characterized in that: The constant current is 0.1C, where C refers to the charge / discharge rate.

4. The lithium-ion battery grouping method based on the random forest algorithm according to claim 2, characterized in that: The fixed time is 10 seconds.

5. A computer system, characterized in that... include: One or more processors, a computer-readable storage medium for storing one or more programs, wherein, when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method of claim 1.

6. A computer-readable storage medium, characterized in that... The device stores computer-executable instructions, which, when executed, are used to implement the method of claim 1.

Images