A method and system for deep sorting of retired batteries

Abstract

The present invention discloses a method and system for deep sorting of retired batteries, which relates to the technical field of battery energy storage. The method includes the following steps: generating multiple original characteristic curves of different types according to sampling data; performing second-order differentiation on the multiple original characteristic curves to obtain multiple velocity characteristic curves and multiple acceleration characteristic curves, respectively obtaining statistical indexes of each curve to obtain multiple baseline characteristics, multiple velocity characteristics, and multiple acceleration characteristics; performing preliminary clustering on the multiple baseline characteristics to obtain an initial grouping result; obtaining the first Euclidean distance and the second Euclidean distance of each retired battery, and re-sorting the retired battery according to the first Euclidean distance and the second Euclidean distance to obtain an optimal grouping result. The present invention significantly expands the feature space through multi-level feature extraction, thereby capturing more complex performance characteristics of the battery to accurately depict the aging degree and future performance of the battery, and realizing precise battery sorting.

Description

Technical Field

[0001] The present invention relates to the technical field of battery energy storage, and particularly to a method and system for deep sorting of retired batteries. Background Art

[0002] With the rapid expansion of the electric vehicle industry, the global demand for lithium-ion batteries has shown an explosive growth trend. According to statistics, the total amount of retired batteries is expected to reach 314 GWh (Gigawatt-hour) in 2030, which is a huge quantity and is expected to account for 15% to 20% of the global energy storage device capacity in the future. Currently, there are mainly two paths for the treatment of retired batteries: direct recycling and secondary utilization. Due to the immaturity of the technology, direct recycling is often accompanied by problems of resource waste and environmental pollution; while secondary utilization realizes the dual improvement of economic and environmental benefits by extending the service life of the battery. It is expected that the global secondary utilization market will create annual revenues of billions of dollars and is expected to reduce the demand for raw materials by 20%.

[0003] Inside the battery, due to the large differences in the usage history, working conditions, and aging degree of each battery cell, the health state of the single cell after retirement is usually in the range of 80% to 90%. This differential characteristic makes it necessary to accurately sort and regroup the single cells before the secondary utilization of retired batteries to avoid the "barrel effect" from having an adverse impact on the long-term performance of the entire battery system.

[0004] Traditional battery sorting methods mainly rely on explicit indicators that can be directly measured, such as capacity, internal resistance, voltage, Coulomb efficiency, and energy efficiency, etc. These indicators are usually obtained through full charge-discharge cycle tests or other test methods. However, these explicit characteristics can only reflect the current state of the battery, and the correlation with the future aging trend and the actual performance after retirement is poor, and it is impossible to accurately sort retired batteries. Although advanced technologies such as electrochemical impedance spectroscopy and X-ray computed tomography can reveal deeper electrochemical information, which helps to achieve more accurate sorting, these methods are costly and complex to operate and are not suitable for large-scale industrial applications. Summary of the Invention

[0005] Based on the defects existing in the above-mentioned prior art, the present invention provides a method and system for deep sorting of retired batteries, which solves the problem that the explicit indicators in traditional sorting methods can only reflect the current state of the battery, and the correlation with the future aging trend and the actual performance after retirement is poor, and it is impossible to accurately sort retired batteries.

[0006] The present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a method for deep sorting of retired batteries, including the following steps:

[0008] Collect the sampling data of each retired battery in a single charging test in real time, and generate multiple original characteristic curves of different types according to the sampling data;

[0009] Derive multiple original characteristic curves to obtain multiple velocity characteristic curves; derive multiple velocity characteristic curves to obtain multiple acceleration characteristic curves; respectively obtain the statistical indexes of multiple original characteristic curves, multiple velocity characteristic curves and multiple acceleration characteristic curves to obtain multiple baseline characteristics, multiple velocity characteristics and multiple acceleration characteristics;

[0010] Cluster multiple baseline feature vectors, and sort multiple retired batteries based on the clustering results to obtain an initial grouping result; among them, the baseline feature vector is a sequence formed by multiple baseline characteristics;

[0011] Obtain the first Euclidean distance between the velocity feature vector of each retired battery in the initial grouping result and its group center, and re-sort the retired battery according to the first Euclidean distance to obtain an optimized grouping result; among them, the velocity feature vector is a sequence formed by multiple velocity characteristics;

[0012] Obtain the second Euclidean distance between the acceleration feature vector of each retired battery in the optimized grouping result and its group center, and re-sort the retired battery according to the second Euclidean distance to obtain an optimal grouping result; among them, the acceleration feature vector is a sequence formed by multiple acceleration characteristics.

[0013] Preferably, the multiple original characteristic curves include a charging capacity-voltage curve in the constant current charging stage, a voltage-charging capacity curve in the constant current charging stage, a current-time curve in the constant voltage charging stage, and a voltage-time curve in the rest stage after charging ends.

[0014] Preferably, the statistical indexes include mean, variance, skewness, kurtosis, and values at the 0%, 25%, 50%, 75%, and 100% quantiles.

[0015] Preferably, the derivation of the multiple original characteristic curves is specifically as follows:

[0016] ;

[0017] In the formula, is the result obtained by the first derivative, represents the ordinate of the original characteristic curve, represents the abscissa of the original characteristic curve, is the change in the abscissa;

[0018] The derivation of the multiple velocity characteristic curves is specifically as follows:

[0019] ;

[0020] In the formula, is the result obtained by the second derivative.

[0021] Preferably, the first Euclidean distance is specifically as follows:

[0022] ;

[0023] In the formula, is the distance from the velocity feature vector to the group center, is the i th velocity feature vector of the battery sample, is the coordinate of the cluster center in the velocity feature space;

[0024] The re-sorting of the retired battery according to the first Euclidean distance specifically includes:

[0025] If there exists a group , its constraint condition is:

[0026] ;

[0027] In the formula, is the i th distance from the velocity feature vector of the sample to the group center, is the first threshold;

[0028] If the first Euclidean distance satisfies the above constraint condition, the corresponding battery sample is reallocated from the current group to a new group .

[0029] Preferably, the second Euclidean distance is specifically as follows:

[0030] ;

[0031] In the formula, is the distance from the acceleration feature vector to the group center, is the i th acceleration feature vector of the battery sample, is the coordinate of the cluster center in the acceleration feature space;

[0032] The re-sorting of the retired battery according to the second Euclidean distance specifically includes:

[0033] If there exists a group , its constraint condition is:

[0034] ;

[0035] Wherein, is the distance from the acceleration eigenvector of the i th sample to the grouping ; is the second threshold;

[0036] If the second Euclidean distance satisfies the above constraints, the corresponding battery sample is reallocated from the current grouping to a new grouping .

[0037] Preferably, before generating multiple original feature curves of different types based on the sampling data, noise filtering, smoothing processing, and interpolation preprocessing need to be performed on the sampling data.

[0038] In a second aspect, the present invention provides a deep sorting system for retired batteries, including:

[0039] An acquisition module, configured to collect sampling data of each retired battery in a single charging test in real time, and generate multiple original feature curves of different types according to the sampling data;

[0040] A derivative module, configured to take the derivative of multiple original feature curves to obtain multiple velocity feature curves; take the derivative of multiple velocity feature curves to obtain multiple acceleration feature curves; respectively obtain statistical indicators of multiple original feature curves, multiple velocity feature curves, and multiple acceleration feature curves to obtain multiple baseline features, multiple velocity features, and multiple acceleration features;

[0041] A clustering module, configured to cluster multiple baseline feature vectors, and sort multiple retired batteries based on the clustering result to obtain an initial grouping result; wherein, the baseline feature vector is a sequence formed by multiple baseline features;

[0042] A first sorting module, configured to obtain the first Euclidean distance between the velocity feature vector of each retired battery in the initial grouping result and the center of its belonging grouping, and re-sort the retired battery according to the first Euclidean distance to obtain an optimized grouping result; wherein, the velocity feature vector is a sequence formed by multiple velocity features;

[0043] A second sorting module, configured to obtain the second Euclidean distance between the acceleration feature vector of each retired battery in the optimized grouping result and the center of its belonging grouping, and re-sort the retired battery according to the second Euclidean distance to obtain an optimal grouping result; wherein, the acceleration feature vector is a sequence formed by multiple acceleration features.

[0044] Compared with the prior art, the above at least one technical solution adopted by the present invention can achieve the following beneficial effects:

[0045] The present invention first collects sampling data of each retired battery in a single charging test in real time, generates an original characteristic curve based on the sampling data, differentiates the original characteristic curve to obtain a velocity characteristic curve; differentiates the velocity characteristic curve to obtain multiple acceleration characteristic curves. Statistical indicators of multiple original characteristic curves, multiple velocity characteristic curves, and multiple acceleration characteristic curves are respectively obtained to obtain multiple baseline characteristics, multiple velocity characteristics, and multiple acceleration characteristics. Through multi-level feature extraction, the present invention significantly expands the feature space, thereby capturing more complex performance characteristics of the battery to accurately characterize the aging degree and future performance of the battery. Then, clustering is performed on multiple baseline feature vectors, and based on the clustering results, multiple retired batteries are sorted to obtain an initial grouping result. Further, by gradually introducing velocity characteristics and acceleration characteristics to adjust the initial grouping result, the performance difference between groups is effectively reduced, and it is ensured that the grouping of the batteries conforms to their aging trend, realizing accurate battery sorting. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0047] Figure 1 It is a flowchart of a method for in-depth sorting of retired batteries according to the present invention;

[0048] Figure 2 It is the original characteristic curve of the charging capacity-voltage curve in the constant current charging stage in the embodiment of the present invention;

[0049] Figure 3 It is the first derivative curve of the charging capacity-voltage curve in the constant current charging stage in the embodiment of the present invention;

[0050] Figure 4 It is the second derivative curve of the charging capacity-voltage curve in the constant current charging stage in the embodiment of the present invention;

[0051] Figure 5 It is the original characteristic curve of the voltage-charging capacity curve in the constant current charging stage in the embodiment of the present invention;

[0052] Figure 6 It is the first derivative curve of the voltage-charging capacity curve in the constant current charging stage in the embodiment of the present invention;

[0053] Figure 7 It is the second derivative curve of the voltage-charging capacity curve in the constant current charging stage in the embodiment of the present invention;

[0054] Figure 8 It is the original characteristic curve of the current-time curve in the constant voltage charging stage in the embodiment of the present invention;

[0055] Figure 9 It is the first derivative curve of the current-time curve in the constant voltage charging stage in the embodiment of the present invention;

[0056] Figure 10 It is the second derivative curve of the current-time curve in the constant voltage charging stage in the embodiment of the present invention;

[0057] Figure 11 It is the original characteristic curve of the voltage-time curve in the static stage after charging in the embodiment of the present invention;

[0058] Figure 12 It is the first derivative curve of the voltage-time curve in the static stage after charging in the embodiment of the present invention;

[0059] Figure 13 It is the second derivative curve of the voltage-time curve in the static stage after charging in the embodiment of the present invention. Specific embodiments

[0060] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0061] To solve the above problems, the present invention discloses a method and system for sorting retired batteries, and its main purpose is to provide a multi-layer feature extraction and deep sorting framework for efficiently and accurately sorting retired batteries, so as to achieve long-term consistency and stability of the overall performance during secondary utilization.

[0062] The present invention provides a method for deep sorting of retired batteries, specifically relating to a multi-layer feature extraction technology and a deep sorting technology based on sampling data. By performing multi-level feature extraction on the data during the charging process of retired batteries, rich implicit features are obtained to accurately characterize the aging degree and future performance of the batteries. The feature extraction technology of the present invention is divided into multiple layers, and each layer processes the static features and dynamic behaviors of the batteries respectively, so as to comprehensively and accurately sort the retired batteries. Refer to Figure 1 , specifically including the following steps:

[0063] S1: Real-time collect the sampling data of each retired battery in a single charging test, and generate multiple original characteristic curves of different types according to the sampling data.

[0064] The sampled data during the charging test includes the data during the charging process and the data during the static stage after charging ends.

[0065] In the present invention, a constant current and constant voltage charging test is performed on multiple retired batteries, and original sampled data such as charging capacity, voltage, current, and temperature are collected in real time.

[0066] First, the collected original sampled data is preprocessed through noise filtering, smoothing, and necessary interpolation. Multiple original feature curves are generated from the preprocessed data. Each original feature curve is standardized into multiple sampling points, and the multiple sampling points of each original feature curve form a data set. This ensures that the data lengths of each curve are the same, facilitating the unified calculation of statistical features.

[0067] The multiple original feature curves include the charge capacity - voltage curve during the constant current charging stage ( ), the voltage - charge capacity curve during the constant current charging stage ( ), the current - time curve during the constant voltage charging stage ( ), and the voltage - time curve during the static stage after charging ends ( ), where is charging capacity is the voltage, is the current, t is the time.

[0068] Furthermore, the four original feature curves can be the complete charging curves of the corresponding charging stages or partial charging segments.

[0069] S2: Respectively obtain the statistical indicators of multiple original feature curves to obtain multiple baseline features; take the derivative of multiple original feature curves to obtain multiple velocity feature curves, respectively obtain the statistical indicators of each velocity feature curve to obtain multiple velocity features; take the derivative of multiple velocity feature curves to obtain multiple acceleration feature curves, and respectively obtain the statistical indicators of each acceleration feature curve to obtain multiple acceleration features.

[0070] First - layer feature extraction: Calculate nine statistical indicators for each original feature curve, and a total of 36 baseline features are obtained.

[0071] Calculate nine statistical indicators for each feature curve at its multiple data points, including the mean, variance, skewness, kurtosis, and the values at the 0%, 25%, 50%, 75%, and 100% quantiles.

[0072] Among them, the mean is used to measure the central position of the data set, and the formula is as follows:

[0073] (1);​

[0074] In the formula, represents the i th data point, N represents the total number of data points, is the mean of the data.

[0075] Variance is used to measure the degree of deviation of data from the mean, and the formula is:

[0076] (2);

[0077] In the formula, is the i th data point, N is the total number of data points, is the variance.

[0078] Skewness is used to describe the symmetry of data distribution, and the formula is:

[0079] (3);

[0080] Among them, is the skewness, is the standard deviation, is the mean, is the i th data point, N is the total number of data points.

[0081] Kurtosis describes the sharpness of data distribution, and the formula is:

[0082] (4);

[0083] Among them, is the kurtosis, is the standard deviation, is the mean, is the i th data point, N is the total number of data points. Subtracting 3 is to make the kurtosis of the normal distribution equal to 0.

[0084] The 0% quantile is the value of the first data point in the dataset, and the formula is:

[0085] (5);

[0086] Among them, is the 0% quantile, is the first data point in the dataset.

[0087] The 25% quantile is the first quartile in a dataset, corresponding to the value of the th data point in the dataset. The formula is:

[0088] (6);

[0089] where is the 25% quantile, represents rounding up, N is the total number of data points.

[0090] The 50% quantile is the median of the dataset, corresponding to the value of the th data point in the dataset. The formula is:

[0091] (7);

[0092] where is the 50% quantile, represents rounding up, N is the total number of data points.

[0093] The 75% quantile is the third quartile in the dataset, corresponding to the value of the th data point in the dataset. The formula is:

[0094] (8);

[0095] where is the 75% quantile, represents rounding up, N is the total number of data points.

[0096] The 100% quantile is the maximum value of the dataset, corresponding to the value of the last data point in the dataset. The formula is:

[0097] (9);

[0098] where is the 100% quantile, [[ID=6�]] is the last data point in the dataset.

[0099] By calculating these statistical indicators for each original feature curve, a total of 36 baseline features are obtained, called the "baseline feature set". These baseline features reflect the static performance during the battery charging process and are the basis for battery performance evaluation.

[0100] Second - layer feature extraction: Derive the four original feature curves to obtain four first - derivative curves:

[0101] (10);

[0102] Among them, represents the ordinate of the original feature curve (e.g., voltage, current, or charging capacity, etc.), represents the abscissa of the original feature curve (such as time and charging capacity, etc.), is a tiny change in the abscissa. By deriving each feature curve once, four first - derivative curves are obtained, which are called "velocity feature curves".

[0103] For each first - derivative curve, continue to calculate nine statistical indicators, including mean, variance, skewness, kurtosis, and values at the 0%, 25%, 50%, 75%, and 100% quantiles. Through this processing, 36 "velocity features" are obtained. These features reflect the rate of change of the corresponding state variables during the battery charging process and can characterize the dynamic characteristics of the battery charging process.

[0104] Third - layer feature extraction: By continuing to derive the four first - derivative curves, four second - derivative curves are obtained:

[0105] (11);

[0106] Among them, is the result obtained from the first - order derivative, is the result obtained from the second - order derivative. By taking the second - order derivative of the first - derivative curves, four second - derivative curves are obtained, which are called "acceleration feature curves".

[0107] For each second - derivative curve, also calculate the values of nine statistical indicators: mean, variance, skewness, kurtosis, and the 0%, 25%, 50%, 75%, and 100% quantiles. Through this processing, 36 acceleration features are obtained. These acceleration features reflect the acceleration effect of the change of state variables during the battery charging process and further capture the hidden dynamic changes of battery behavior.

[0108] Through the three - layer feature extraction of the present invention, each retired battery obtains 108 latent features, including 36 baseline features, 36 velocity features, and 36 acceleration features. These features describe the static attributes and dynamic behaviors of retired batteries from multiple dimensions, forming a hierarchical feature representation.

[0109] S3: Adopt a depth - sorting framework based on the derivative - increment improvement mechanism and use the 108 latent features for grouping retired batteries.

[0110] The present invention provides a depth sorting framework based on multi-layer hidden features. This framework optimizes the battery grouping by gradually introducing speed features and acceleration features, making the grouping of each battery more accurate and stable, and better reflecting the aging behavior of the battery. The sorting includes three stages: initial grouping, first-stage refinement, and second-stage refinement. Among them, in the initial grouping, baseline features are used to preliminarily group battery samples. In the first-stage refinement, the initial grouping is fine-tuned based on the extracted speed features. In the second-stage refinement, acceleration features are further introduced to fine-tune the battery grouping.

[0111] In the initial grouping process, the K-means clustering algorithm is used, and the clustering centers are initialized by the k-means++ method to avoid poor initialization and reduce the risk of local minima. The K-means method first randomly selects a sample point as the initial clustering center, and then selects subsequent centers according to the probability of the squared distance from the existing centers, ensuring that the clustering centers are evenly distributed in the feature space. The formula is as follows:

[0112] (12);

[0113] Among them, is the baseline feature vector of the i th battery sample, is the center of the cluster to which the sample belongs, N 1 is the total number of battery samples, is the cluster number to which the sample belongs. The K-means algorithm updates the grouping of samples and the clustering centers through multiple iterations until the assignment changes slightly or reaches the convergence condition.

[0114] Based on the extracted speed features, an incremental improvement mechanism based on derivatives is used to fine-tune the initial grouping. Specifically, for each battery sample, calculate its alignment distance (the first Euclidean distance) in the speed feature space:

[0115] (13);

[0116] Among them, is the speed feature vector of the i th battery sample, is the coordinate of the cluster center in the speed feature space, is the distance from the speed feature vector to the grouping center.

[0117] If there exists a grouping , which satisfies:

[0118] (14);

[0119] Among them, is the distance from the velocity feature vector of the i th sample to the grouping center, is the first threshold defined by the user, representing the strictness of sample reallocation. If the condition is met, the sample is reallocated from the current grouping to a new grouping . This step effectively optimizes the subtle differences in the baseline feature grouping by introducing velocity features, ensuring the accuracy of retired battery sorting and recombination.

[0120] Further introduce acceleration features to fine-tune the battery grouping. For each battery sample, calculate its alignment distance (the second Euclidean distance) in the acceleration feature space:

[0121] (15);

[0122] Among them, is the distance from the acceleration feature vector to the grouping center, is the i th acceleration feature vector of the battery sample, is the coordinate of the clustering center in the acceleration feature space.

[0123] If there exists a grouping , and its distance satisfies:

[0124] (16);

[0125] In the formula, is the distance from the acceleration feature vector of the i th sample to the grouping , is the second threshold defined by the user, representing the strictness of sample reallocation. If the condition is met, the sample is reallocated to this grouping . Among them, this step further reveals the aging mode of the battery by introducing acceleration features and makes a finer adjustment to the grouping.

[0126] Embodiment 1

[0127] Through long-term aging tests on 86 retired lithium-ion batteries, an experimental dataset containing battery sampling data is constructed, and data preprocessing, multi-layer feature extraction, and a deep sorting framework are used to achieve precise grouping of retired batteries. The overall schematic diagram is as shown in Figure 1 , including the following steps:

[0128] S4: Collect the aging and single-charge test data of 86 lithium-ion batteries through battery experiments, and preprocess the collected original sampling data.

[0129] Select 86 nickel-cobalt-manganese / graphite lithium-ion batteries (nominal capacity 2.4 Ah, nominal voltage 3.7 V). Among them, the experiment is divided into two stages:

[0130] The first life cycle: Divide the 86 batteries into 16 groups according to 16 charge-discharge protocols. After multiple cycles, the battery capacity decays from 100% (2.4 Ah) to 80% (1.92 Ah).

[0131] The second life cycle: All retired batteries are uniformly cycled and aged under the conditions of charging at 0.5C and discharging at 0.2C until the capacity drops to 50% (1.2 Ah).

[0132] During each charging test, sampling data is collected in real time to obtain a large amount of charging curve data, providing a basis for subsequent feature extraction.

[0133] The collected original sampling data is subjected to noise filtering, smoothing processing, and necessary interpolation processing, and each key feature curve is standardized to 128 sampling points. The obtained 128 data points can better depict the overall trend of the curve during the charging process, providing high-quality data for subsequent multi-layer feature extraction. For the feature curve and its derivative curve, see Figure 2-13 .

[0134] S5: Adopt a three-layer feature engineering strategy to extract a total of 108 implicit features from the four key feature curves and their derivative curves.

[0135] The first layer of feature extraction (baseline features): Using the preprocessed and interpolated charging curve, obtain the following four feature curves: the charge capacity-voltage curve during the constant current charging stage ( ), the voltage-charge capacity curve during the constant current charging stage ( ), the current-time curve during the constant voltage charging stage ( ), and the voltage-time curve during the rest stage after charging ends ( ).

[0136] Directly calculate nine statistical indicators for each curve at its 128 data points: mean, variance, skewness, kurtosis, and the values at 0%, 25%, 50%, 75%, and 100%. Thus, 9 indicators for each of the four curves are obtained, generating a total of 36 baseline features.

[0137] The second layer of feature extraction (speed features): Take the first derivative of each of the above four original feature curves to obtain the first derivative curves. Calculate the above nine statistical indicators for each first derivative curve to obtain 36 speed features, which reflect the rate of change of each parameter during the battery charging process (see the corresponding speed curves in Figure 2 ).

[0138] Third-layer feature extraction (acceleration features): Based on the first-order derivative curve, a second-order derivative curve is derived. The same nine statistical indicators are calculated for each second-order derivative curve, resulting in 36 acceleration features.

[0139] Combining the above three layers, a total of 108 implicit features (36 baselines + 36 speeds + 36 accelerations) are obtained, which constitute a hierarchical description of battery charging behavior.

[0140] S6: A deep sorting framework based on derivative incremental improvement mechanism is adopted to use 108 latent features for grouping retired batteries.

[0141] Initial Grouping (Baseline Feature Guided): Using the 36 baseline features extracted above, the 86 retired batteries were initially grouped using the K-means clustering algorithm. In this initial phase, cluster centers were determined using the K-means++ method, ensuring a uniform distribution in the feature space. The batteries were initially grouped into eight groups. This initial grouping provided a stable foundation for subsequent fine-tuning. The statistical results from this initial phase showed the following:

[0142] The average aging difference is about 61.72.

[0143] The average capacity utilization rate is approximately 67.97%.

[0144] The average service life is about 285 cycles.

[0145] First-stage refinement (speed feature guidance): Based on the initial grouping, speed features are used to fine-tune some batteries. For each battery, the Euclidean distance between its eigenvector in the speed feature space and the cluster center of its group is calculated; its distance to the cluster centers of all other groups is also calculated.

[0146] After the first phase of adjustments, the evaluation indicators of each group improved to:

[0147] The aging difference drops to about 45.27.

[0148] Capacity utilization increased to approximately 75.38%.

[0149] The average service life is extended to approximately 321 cycles.

[0150] The second stage of refinement (acceleration feature guidance): further introduce the acceleration feature to fine-tune the grouping results, and use a similar fine-tuning strategy to redistribute some batteries that are closer to other groups in the acceleration feature space. After the second stage of adjustment, the final grouping results are as follows:

[0151] The mean aging difference was reduced to 44.62.

[0152] The average capacity utilization rate is increased to 78.06%.

[0153] The average service life is extended to 334 cycles.

[0154] S7: Compare the deep sorting method of the present invention with the traditional sorting method based on explicit features.

[0155] To verify the advantages of the deep sorting method of the present invention, a traditional sorting method based on explicit features (such as capacity, internal resistance, voltage, and coulomb efficiency energy efficiency) is compared using the same dataset. The comparison results show that:

[0156] Aging difference: The grouping aging difference of the traditional explicit feature method is usually above 75, while the aging difference of the final grouping of the present invention is reduced to about 44.62, a decrease of about 40.9%.

[0157] Capacity utilization rate: The average capacity utilization rate in the traditional grouping is about 61.59%, while this method is increased to 78.06%, an increase of about 26.7%.

[0158] Average service life: The average cycle life of each group under the traditional method is about 253 cycles, and the method of the present invention is extended to 334 cycles, an increase of about 32.0%.

[0159] The above comparison results fully prove that the deep sorting method based on 108 - dimensional implicit features and phased refined adjustment has significant advantages in improving the long - term consistency and secondary utilization performance of retired batteries.

[0160] This embodiment details how to use the single - charge test data of 86 retired batteries to achieve precise grouping of retired batteries through data pre - processing, multi - level feature extraction, and a grouping method based on a deep sorting framework. By introducing speed and acceleration features for fine - tuning, this method not only significantly reduces the within - group aging difference, improves the capacity utilization rate, and extends the average service life of each group of batteries, but also shows obvious superiority in comparison with the traditional sorting method based on explicit features.

[0161] In addition, due to the lack of a public, systematic, and long - term tracked dataset of retired batteries, the prior art faces great difficulties in long - term verification of the performance of sorted battery packs, further exacerbating the development obstacles of retired battery reuse technology.

[0162] The present invention provides a multi - layer feature extraction technique, which can comprehensively and accurately characterize the performance state of retired batteries. Through multi - level feature extraction, the feature space is significantly expanded, thereby capturing more complex performance characteristics of the batteries; this method can not only evaluate the current state of the batteries, but also accurately reflect the dynamic changes of the batteries during the charging process; this method does not rely on the historical usage data of retired batteries, reducing the complexity of data collection and processing.

[0163] The present invention provides a deep sorting framework that can achieve precise battery grouping. By combining different levels of implicit features, it can effectively capture the subtle differences of batteries. Further, by gradually introducing speed features and acceleration features, the battery grouping is refined. The high efficiency and stability of the sorting process make it suitable for the secondary utilization of large - scale retired batteries.

[0164] Based on the same concept, the present invention also provides a deep sorting system for retired batteries, which can automatically complete the processes of data collection, feature extraction, deep sorting, and result display of retired batteries. By integrating multiple modules and combining multi - level feature extraction and the deep sorting framework, the high - precision and high - efficiency of battery sorting are ensured. Through the combination of hardware and software systems, the whole - process automation of retired batteries from data collection, feature extraction to deep sorting is realized. The system includes a collection module, a derivative module, a clustering module, a first sorting module, and a second sorting module.

[0165] The collection module is used to collect sampling data of each retired battery in a single charging test in real - time, and generate multiple original feature curves of different types according to the sampling data.

[0166] The derivative module is used to take the derivative of multiple original feature curves to obtain multiple speed feature curves; take the derivative of multiple speed feature curves to obtain multiple acceleration feature curves; respectively obtain the statistical indexes of multiple original feature curves, multiple speed feature curves, and multiple acceleration feature curves to obtain multiple baseline features, multiple speed features, and multiple acceleration features.

[0167] The clustering module is used to cluster multiple baseline feature vectors, and sort multiple retired batteries based on the clustering result to obtain an initial grouping result; among them, the baseline feature vector is a sequence formed by multiple baseline features.

[0168] The first sorting module is used to obtain the first Euclidean distance between the speed feature vector of each retired battery in the initial grouping result and the group center to which it belongs, and re - sort the retired battery according to the first Euclidean distance to obtain an optimized grouping result; among them, the speed feature vector is a sequence formed by multiple speed features.

[0169] The second sorting module is used to obtain the second Euclidean distance between the acceleration feature vector of each retired battery in the optimized grouping result and its belonging grouping center, and re-sort the retired battery according to the second Euclidean distance to obtain the optimal grouping result; wherein, the acceleration feature vector is a sequence formed by multiple acceleration features.

[0170] Among them, the acquisition module is used to conduct a single charging test on the retired battery and collect original data such as voltage, current, and temperature in real time. This module is connected to the battery charging device through sensors, can accurately record the dynamic changes of the battery during charging, and transmit this original data to the subsequent processing unit of the system. The functions of the data acquisition module include:

[0171] Collect the voltage, current, and temperature data of the battery charging process in real time.

[0172] Support multi-channel data synchronous acquisition to ensure data integrity.

[0173] Monitor the battery status in real time through sensors and capture any abnormal behaviors.

[0174] This module provides the real-time status data of the retired battery for the system and is the basis for subsequent feature extraction and sorting processing.

[0175] The derivative module is a multi-layer feature extraction module. The multi-layer feature extraction module is responsible for extracting high-dimensional implicit features from the collected data and constructing the feature vector of the battery. This module includes:

[0176] The preprocessing unit: used to perform preprocessing operations such as denoising, smoothing, and interpolation on the original data to ensure data quality.

[0177] The baseline feature extraction unit: extracts the first-layer baseline features from the original feature curve, such as mean, variance, skewness, kurtosis, and quartiles, and generates 36 baseline features.

[0178] The speed feature extraction unit: performs a first derivative on the original feature curve to obtain the first derivative curve, extracts the change rate features during the battery charging process, and forms 36 speed features.

[0179] The acceleration feature extraction unit: performs a second derivative on the first derivative curve to extract the acceleration features during the battery charging process and generates 36 acceleration features.

[0180] Through these multi-level feature extractions, this module can generate a high-dimensional feature vector containing a large number of implicit features, so as to accurately describe the aging behavior and performance status of the battery.

[0181] The clustering module uses the baseline features and the K-means clustering algorithm to conduct a preliminary grouping of the retired batteries to ensure the stability and accuracy of the preliminary grouping.

[0182] The first sorting module, based on speed features, introduces a fine-tuning mechanism to refine the initial grouping, so as to optimize the boundaries of battery grouping. An incremental improvement mechanism using derivative features is adopted to reassign only those samples that are close to other groups in the speed feature space.

[0183] The second sorting module further fine-tunes the grouping results based on acceleration features to further optimize the stability and consistency of the grouping, ensure the maximum similarity of battery performance, and ensure a high degree of uniformity in the battery aging behavior within each group.

[0184] Through this series of refined adjustments, the performance differences between groups can be effectively reduced, and the grouping of batteries can be ensured to conform to their aging trends.

[0185] The system is integrated in a dedicated control chip or computer to achieve a fast and accurate grouping process. The system can use an industrial-grade computing platform or an embedded processor in terms of hardware, and combine dedicated sensors and data acquisition devices to achieve real-time data acquisition; the software part integrates functions such as data preprocessing, feature extraction, and in-depth sorting.

[0186] The entire process completed by the system is as follows: (1) Sampling data acquisition for a single charge test, (2) Sampling data preprocessing, (3) Multi-layer feature extraction of curves, (4) Initial grouping based on baseline features (baseline clustering), (5) First-stage refinement based on speed features (speed feature fine-tuning), (6) Second-stage refinement based on acceleration features (acceleration feature re-optimization), (7) Obtaining sorting results.

[0187] After system integration, the entire sorting process of retired batteries realizes fully automatic operation, which not only greatly improves the sorting efficiency, but also ensures the grouping accuracy, providing reliable and economical technical support for the secondary utilization of batteries.

[0188] This system can not only overcome the limitations of existing explicit feature methods in predicting future performance, but also achieve accurate grouping of retired batteries on the premise of low cost and easy operation, providing a reliable basic guarantee for secondary utilization.

[0189] The present invention provides a system for in-depth sorting of retired batteries. The system can complete the feature extraction and sorting processing of batteries in a short time, greatly improving the processing efficiency; the entire system realizes automatic processing from data acquisition, feature extraction to sorting optimization, reducing manual intervention and lowering the operation difficulty; in addition, the system has good scalability and can adapt to different scales of retired battery sorting tasks to meet industrial needs.

[0190] Although the preferred embodiments of the present invention have been described, additional changes and modifications can be made to these embodiments by those skilled in the art once they learn the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present invention.

[0191] Obviously, those skilled in the art can make various changes and deformations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and deformations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and deformations.

Claims

1. A method for in-depth sorting of retired batteries, characterized in that, Including the following steps: Collecting the sampling data of each retired battery in a single charging test in real time, and generating multiple original feature curves of different types according to the sampling data; Taking the derivative of multiple original feature curves to obtain multiple velocity feature curves; taking the derivative of multiple velocity feature curves to obtain multiple acceleration feature curves; respectively obtaining the statistical indexes of multiple original feature curves, multiple velocity feature curves and multiple acceleration feature curves to obtain multiple baseline features, multiple velocity features and multiple acceleration features; Clustering multiple baseline feature vectors, and sorting multiple retired batteries based on the clustering result to obtain an initial grouping result; wherein, the baseline feature vector is a sequence formed by multiple baseline features; Obtaining the first Euclidean distance between the velocity feature vector of each retired battery in the initial grouping result and its group center, and re-sorting the retired battery according to the first Euclidean distance to obtain an optimized grouping result; wherein, the velocity feature vector is a sequence formed by multiple velocity features; Obtaining the second Euclidean distance between the acceleration feature vector of each retired battery in the optimized grouping result and its group center, and re-sorting the retired battery according to the second Euclidean distance to obtain an optimal grouping result; wherein, the acceleration feature vector is a sequence formed by multiple acceleration features.

2. The method for deep sorting of retired batteries according to claim 1, wherein The multiple original feature curves include a charge capacity-voltage curve in the constant current charging stage, a voltage-charge capacity curve in the constant current charging stage, a current-time curve in the constant voltage charging stage, and a voltage-time curve in the rest stage after charging ends.

3. The deep sorting method for retired batteries according to claim 1, characterized in that The statistical indexes include mean, variance, skewness, kurtosis, and values at the 0%, 25%, 50%, 75%, and 100% quantiles.

4. The method for deep sorting of retired batteries according to claim 1, wherein The formula for taking the derivative of multiple original feature curves is specifically as follows: ； In the formula, is the result obtained from the first derivative, represents the ordinate of the original characteristic curve, represents the abscissa of the original characteristic curve, is the change in the abscissa; The formula for taking the derivative of multiple velocity feature curves is specifically as follows: ； In the formula, is the result obtained by the second derivative.

5. A method for deep sorting of retired batteries according to claim 1, characterized in that, The first Euclidean distance is specifically as follows: ； wherein, is the distance from the velocity feature vector to the group center, is the i th velocity feature vector of the battery sample, are the coordinates of the clustering center in the velocity feature space; The re-sorting of the retired battery according to the first Euclidean distance specifically includes: If there exists a group , and its constraint conditions are: ； In the formula, is the distance from the velocity feature vector of the i th sample to the grouping center, and is the first threshold; If the first Euclidean distance satisfies the above constraint conditions, reassign the corresponding battery sample from the current group to a new group .

6. The method for deep sorting of retired batteries according to claim 1, characterized in that, The second Euclidean distance is specifically as follows: ； Wherein, is the distance from the acceleration feature vector to the group center, is the i acceleration feature vector of the th battery sample, and is the coordinate of the clustering center in the acceleration feature space; The re-sorting of the retired battery according to the second Euclidean distance specifically includes: If there exists a group , and its constraint conditions are: ； In the formula, is the distance from the acceleration feature vector of the i th sample to the grouping , and is the second threshold; If the second Euclidean distance satisfies the above constraints, reassign the corresponding battery sample from the current group to a new group .

7. The method for deep sorting of retired batteries according to claim 1, characterized in that Before generating multiple original feature curves of different types according to the sampling data, it is necessary to perform noise filtering, smoothing processing and interpolation preprocessing on the sampling data.

8. A deep sorting system for retired batteries, characterized in that, Including: A collection module for collecting the sampling data of each retired battery in a single charging test in real time, and generating multiple original feature curves of different types according to the sampling data; A derivative module for taking the derivative of multiple original feature curves to obtain multiple velocity feature curves; taking the derivative of multiple velocity feature curves to obtain multiple acceleration feature curves; respectively obtaining the statistical indexes of multiple original feature curves, multiple velocity feature curves and multiple acceleration feature curves to obtain multiple baseline features, multiple velocity features and multiple acceleration features; A clustering module for clustering multiple baseline feature vectors, and sorting multiple retired batteries based on the clustering result to obtain an initial grouping result; wherein, the baseline feature vector is a sequence formed by multiple baseline features; The first sorting module is used to obtain the first Euclidean distance between the speed feature vector of each retired battery in the initial grouping result and its affiliated grouping center, and re-sort the retired battery according to the first Euclidean distance to obtain an optimized grouping result; wherein, the speed feature vector is a sequence formed by a plurality of speed features. The second sorting module is used to obtain the second Euclidean distance between the acceleration feature vector of each retired battery in the optimized grouping result and its affiliated grouping center, and re-sort the retired battery according to the second Euclidean distance to obtain an optimal grouping result; wherein, the acceleration feature vector is a sequence formed by a plurality of acceleration features.

Images