A method for lithium battery dynamic sorting and matching

Abstract

The application provides a lithium battery dynamic sorting and grouping method, which comprises the following steps: obtaining charge and discharge data in a formation stage of the battery, calculating an incremental capacity curve by using a LEAN method, and performing noise reduction processing by using a RGSF multi-iteration Gaussian filter to extract a sorting feature vector; performing three-way verification by using a physical threshold pre-screening, an isolated forest algorithm and a Mahalanobis distance to detect and eliminate abnormal batteries; performing standardization processing on the feature vector of normal batteries, performing soft assignment grouping by using a fuzzy C-means clustering algorithm, and marking boundary batteries; extracting six-dimensional consistency evaluation indexes, calculating a curve distance between batteries in the same gear by using a dynamic time warping algorithm, and outputting a comprehensive consistency score. The application reduces IC curve noise by using multi-iteration filtering, comprehensively identifies abnormal batteries by three-way verification, avoids boundary oscillation problems by using FCM soft clustering, realizes comprehensive consistency evaluation based on a six-index system, and improves the precision and stability of battery sorting and grouping.

Description

Technical Field

[0001] This invention relates to the field of battery evaluation and screening technology, and in particular to a method for dynamically sorting and grouping lithium batteries. Background Technology

[0002] The performance and cycle life of lithium-ion battery packs are primarily constrained by the inconsistencies between individual battery cells. Studies have shown that an initial 20% difference in internal resistance among cells in the same batch can lead to approximately a 40% reduction in cycle life, and cells with an initial parameter dispersion of less than 1% can exhibit a capacity dispersion exceeding 10% after 1000 cycles. Therefore, precise sorting and grouping of cells before battery module assembly is a crucial process for improving the overall performance and extending the lifespan of the battery pack.

[0003] Currently, the sorting technologies for lithium-ion batteries mainly include the following methods: The first method is capacity sorting, which categorizes batteries solely based on their discharge capacity. This method is simple to implement, but because it only considers a single static parameter, it cannot reflect the differences in the dynamic charge and discharge characteristics of the batteries, resulting in significant inconsistency degradation of the sorted battery packs during cycling.

[0004] The second method is a multi-parameter sorting method based on hard clustering algorithms such as K-means. Although this method considers multi-dimensional features, it uses hard boundary partitioning, which makes battery grouping near the cluster boundaries unstable, prone to grouping oscillations, and sensitive to the initialization of cluster centers.

[0005] The third method is based on electrochemical impedance spectroscopy (EIS). This method assesses the battery state by measuring the battery's impedance response at different frequencies, providing rich electrochemical information. However, EIS testing typically requires more than 10 minutes of testing time and necessitates dedicated impedance testing equipment, resulting in high equipment costs and making it unsuitable for rapid sorting applications at the production line level.

[0006] The fourth type is anomaly detection based on a single statistical method, such as the 3sigma rule or a single machine learning method. These methods often can only detect specific types of anomalies and cannot fully cover data anomalies caused by process interruptions, statistical outliers, and distribution deviations, which can easily lead to false positives or false negatives.

[0007] In addition, existing consistency evaluation methods usually only focus on static capacity consistency, lack a systematic evaluation of the similarity of battery dynamic characteristics, and the evaluation index system is incomplete, making it difficult to fully reflect the actual consistency level of battery packs.

[0008] Therefore, there is an urgent need for a sorting and grouping method that can effectively reduce IC curve noise, comprehensively identify abnormal batteries, achieve soft boundary grouping, and systematically evaluate battery consistency, so as to improve the overall performance and cycle life of battery packs. Summary of the Invention

[0009] To achieve the above objectives, the present invention adopts the following technical solution: A method for dynamically sorting and grouping lithium batteries includes the following steps: S1. Obtain charging and discharging data during the battery formation stage, calculate incremental capacity curves for voltage-current data during the charging stage, and perform noise reduction processing using multi-iteration Gaussian filtering to extract sorting feature vectors. S2, triple-verification abnormal battery detection, detects and removes abnormal batteries; S3. The sorting feature vectors of normal batteries are standardized, and fuzzy C-means clustering algorithm is used for soft grouping and boundary batteries are marked. S4. Extract consistency evaluation indicators, use dynamic time warping algorithm to calculate the curve distance between batteries of the same grade, and output comprehensive consistency score.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention uses RGSF multi-iteration Gaussian filtering to process incremental capacity curves. Compared with the traditional single-filtering method, the noise RMSE is reduced by 6-29%, which effectively improves the extraction accuracy of IC curve features. (2) The present invention adopts a triple verification strategy of physical threshold pre-screening, isolated forest statistical detection and Mahalanobis distance fine screening, which can comprehensively identify different types of abnormal batteries such as process abnormalities, statistical abnormalities and distribution abnormalities, thereby improving the accuracy and completeness of abnormality detection. (3) The present invention adopts a soft assignment mechanism of fuzzy C-means clustering, which explicitly reflects the determinism of battery affiliation through membership degree and marks boundary batteries, thus avoiding the boundary oscillation problem of hard clustering method and improving the stability of grouping. (4) This invention establishes a six-index evaluation system and introduces DTW curve distance, covering capacity, voltage and dynamic characteristics, and realizes a comprehensive and systematic evaluation of battery consistency; (5) This invention extracts features based on conventional formation charge and discharge data, without the need for complex testing equipment such as EIS, and is suitable for rapid sorting applications at the production line level. Attached Figure Description

[0011] Figure 1 This is a flowchart of a dynamic sorting and grouping method for lithium iron phosphate batteries based on incremental capacity curve characteristics, according to an embodiment of the present invention. Figure 2This is a schematic diagram of incremental capacity curve feature extraction according to an embodiment of the present invention; Figure 3 This is a flowchart of the triple anomaly detection method according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the FCM clustering membership distribution in an embodiment of the present invention; Figure 5 This is a schematic diagram of the consistency evaluation system of indicators in Embodiment Six of the present invention; Figure 6 This is a bar chart comparing the consistency scores of different sorting methods in embodiments of the present invention. Detailed Implementation

[0012] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0013] like Figure 1-6 As shown, a method for dynamically sorting and grouping lithium batteries includes the following steps: S1: IC curve feature extraction enhanced by multi-iteration Gaussian filtering; S2: Triple-verification abnormal battery detection; S3: Soft allocation and sorting based on FCM fuzzy clustering; S4: Consistency evaluation based on six indicators of DTW.

[0014] like Figure 1-2 As shown, the specific implementation of step S1 is as follows: First, charge and discharge data during the battery formation stage are acquired, including voltage sequence V(t), current sequence I(t), and time sequence t. The sampling interval dt is set to 10 seconds.

[0015] The incremental capacity curve is calculated using the LEAN method. The LEAN method estimates the dQ / dV value by counting the number of sampling points falling within a specific voltage range. The calculation formula is as follows:

[0016] Where: dQ / dV is the differential capacity (incremental capacity), representing the capacity change corresponding to a unit voltage change, usually expressed in Ah / V; V i For the current voltage point being analyzed, the formula gives the differential capacitance value at that voltage, N. i To fall within the voltage range [V i -dV / 2,V i The number of sampling points is I[+dV / 2]. avgThe average current over the time interval dt is expressed in A, where dt is a time element in seconds, and dV is the voltage resolution, set to 4mV.

[0017] The calculated incremental capacity curve is processed using a multi-iteration Gaussian filtering method (RGSF) to reduce noise. The Gaussian filtering function is as follows:

[0018] Where x represents the incremental capacity (dQ / dV), G(x) represents the noise distribution of the measured value, the Gaussian kernel parameter sigma is set to 2.0, and the number of iterations is set to 3. Compared with the traditional single-step filtering method, three-step filtering can reduce the noise RMSE by 6-29%.

[0019] like Figure 3 As shown: The specific implementation of step S2 is as follows: A triple verification strategy is used to detect abnormal batteries: First layer: Physical threshold pre-screening. Based on battery process parameters and expert experience, reasonable physical ranges for each characteristic are set. For example: V m For values ​​in the range [3.38, 3.42]V, dQ / dV m Batteries belonging to [2500, 3500] Ah / V, and ΔQ belonging to [330, 340] Ah, etc., are marked as physical anomalies if they fall outside these ranges.

[0020] The second layer: Isolation Forest statistical detection. For batteries that passed the first layer's detection, the Isolation Forest algorithm is used for statistical anomaly detection. This algorithm is based on the principle that outliers are more easily isolated. It constructs isolation trees by randomly partitioning the data and calculates the average path length required for an outlier to be isolated. The contamination rate parameter is set to 0.02.

[0021] Third layer: Mahalanobis distance fine screening. Calculate the Mahalanobis distance of each cell to the center of the normal cell group:

[0022] Where x is the feature vector of the battery to be tested, u is the mean vector of the normal battery group, and S −1 This is the inverse of the covariance matrix; the threshold is determined based on the 95th quantile of the chi-square distribution.

[0023] like Figure 4 As shown, the specific implementation of step S3 is as follows: First, the sorting feature vectors of normal batteries are standardized using Z-score:

[0024] Where z is the standardized value, x is the original data to be standardized, u is the mean of the feature in the normal battery population, and s is the standard deviation of the feature in the normal battery population.

[0025] Fuzzy C-means (FCM) clustering algorithm was used for grouping. The number of clusters K was set to 4 (corresponding to 4 levels), and the fuzzy coefficient m was set to 1.5. The core of the FCM algorithm is to calculate the membership matrix and update the cluster centers. The membership calculation formula is:

[0026] Among them, u ij Let d be the membership degree of the i-th data point to the j-th cluster center. ij Let d be the distance from the i-th data point to the j-th cluster center. ik Let be the distance from the i-th data point to the k-th cluster center.

[0027] The formula for updating cluster centers is:

[0029] Iterate until convergence, and determine the level to which each battery belongs based on the maximum membership degree principle. Batteries with a maximum membership degree less than 0.6 are marked as edge batteries.

[0030] like Figure 5 As shown, the specific implementation of step S4 is as follows: Extracting six-dimensional evaluation indicators: dQdV m (IC peak value), V_dQdV m (Peak voltage), Volt_start (starting voltage), Volt_end (ending voltage), Rate_CC_CV (CC / CV ratio), a_Ci (curve distance).

[0031] For batteries within the same battery category, calculate the DTW distance between each pair of charging voltage curves:

[0033] Calculate the standard deviation of each indicator within each level and normalize it:

[0035] Final calculation of the overall consistency score:

[0037] Among them, std j std is the original standard deviation of the j-th feature (or sample); min The minimum of the standard deviations of all features (or samples); std max The maximum value among the standard deviations of all features (or samples); std norm,j is the normalized standard deviation of the j-th feature (or sample).

[0038] A higher PC value indicates better consistency.

[0039] Experimental verification: This embodiment uses the formation data of 3546 106Ah LFP cells provided by a battery manufacturer for verification.

[0040] The abnormal detection results are shown below. Figure 3 .

[0041] See the comparison of sorting results. Figure 6 .

[0042] The experimental results show that the FCM dynamic sorting method of this invention improves the consistency score by 15.2% compared with the traditional capacity sorting method and by 5.7% compared with the K-means method. The coefficient of variation (CV) of capacity at each level is controlled within 0.7%, which meets the requirements of production line-level sorting.

[0043] The above are merely optional embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for dynamically sorting and grouping lithium batteries, characterized in that, Includes the following steps: S1. IC curve feature extraction enhanced by multi-iteration Gaussian filtering; S2, triple-verification abnormal battery detection; S3. Soft allocation and sorting based on FCM fuzzy clustering; S4. Consistency evaluation of six indicators based on DTW.

2. The method for dynamically sorting and grouping lithium batteries according to claim 1, characterized in that, Step S1 includes the following steps: S11. First, obtain the charge and discharge data during the battery formation stage; S12. Calculate the incremental capacity curve using the LEAN method; S13. Perform RGSF multi-iteration Gaussian filtering on the calculated incremental capacity curve to reduce the impact of noise.

3. The method for dynamically sorting and grouping lithium batteries according to claim 2, characterized in that, In step S11, the charge and discharge data includes the charging and discharging time, voltage, and current during the battery formation stage.

4. The method for dynamically sorting and grouping lithium batteries according to claim 1, characterized in that, In step S2, the triple verification includes physical threshold, isolated forest and Mahalanobis distance verification.

5. The method for dynamically sorting and grouping lithium batteries according to claim 1, characterized in that, Step S3 includes the following steps: S31. Perform Z-score standardization on the sorting feature vector of normal batteries; S32. Grouping is performed using the fuzzy C-means clustering algorithm; S33. Iterate until convergence, and determine the gear level to which each battery belongs based on the principle of maximum membership.

6. The method for dynamically sorting and grouping lithium batteries according to claim 4, characterized in that, Step S32 includes calculating the membership matrix and updating the cluster centers.

7. The method for dynamically sorting and grouping lithium batteries according to claim 1, characterized in that, Step S4 includes the following steps: S41. Extract six-dimensional evaluation indicators; S42. For batteries in the same category, calculate the DTW distance between the charging voltage curves of each pair. S43. Calculate the standard deviation of each indicator within each gear level and normalize it; S44. Calculate the final overall consistency score.

8. The method for dynamically sorting and grouping lithium batteries according to claim 7, characterized in that, In step S41, the evaluation metrics include IC peak value, peak voltage, starting voltage, ending voltage, CC / CV ratio, and curve distance.

Images