Rapid sorting method and device for echelon utilization batteries and storage medium

Abstract

The invention discloses a rapid sorting method and device for echelon utilization batteries and a storage medium, and relates to the technical field of echelon utilization batteries. In order to overcome the defects that an existing retired battery sorting method is inaccurate in sorting, low in efficiency, high in equipment requirement and incapable of achieving large-scale batch sorting, a battery sample is charged to preset cut-off voltage, and standing testing is conducted; recording a standing voltage sequence; carrying out feature extraction on the collected standing voltage sequence, and constructing a sorting feature vector required by battery sorting; dynamic modulation and gating weighted fusion are carried out on the feature vector obtained through wavelet decomposition and the cut-off voltage; classifying the performance grades of the batteries according to the SOH; and performing model training based on the modulation fusion features and the performance grades, obtaining feature vectors and cut-off voltages of the to-be-sorted echelon utilization batteries, and inputting the trained battery sorting model to obtain performance grade prediction results of the batteries. The method is mainly used for rapidly sorting large-scale retired batteries.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cascade utilization battery, and particularly relates to a rapid sorting method and device for cascade utilization battery and a storage medium. BACKGROUND

[0002] After a certain period of use, the capacity of lithium batteries decreases, the internal resistance increases, and the safety decreases, which is difficult to meet the use requirements of the power system, and must be replaced and retired. The types of batteries entering the retirement stage are complex, and the performance differences are significant. If they are directly scrapped, not only resources are wasted, but also environmental pollution and safety risks exist. A considerable part of the retired batteries still has high residual capacity and cycle performance, which can be used for energy storage, communication base stations, electric bicycles and other cascade utilization scenarios. In order to maximize resource recovery and green recycling, it is necessary to efficiently and accurately sort the performance of retired batteries, and classify them according to their state of health (SOH) and remaining useful life (RuL) for cascade utilization. This not only relates to the safety and stability of the cascade battery application system, but also directly affects the economic benefits and recycling value.

[0003] Due to the complex sources and numerous types of retired batteries, the original data in their operation process are often missing or not in a unified format, making it difficult to build a standardized state evaluation system. At the same time, the state of the battery near the retirement stage is highly unstable, and the capacity or SOH estimated by the vehicle-mounted BMS (Battery Management System) has a large error, which affects the accuracy and reliability of the screening and grading. Therefore, the evaluation method relying on complete historical data or embedded sensing system has limited adaptability in actual cascade utilization scenarios, and an accurate grading method for retired batteries under the condition of no complete data is urgently needed.

[0004] The current retired battery sorting method mainly includes manual measurement method, cycle test method, model estimation method, etc. The manual experience method relies on manual visual inspection, voltage test and other means for rough judgment, but cannot effectively identify the internal health state of the battery, has poor accuracy and low efficiency, and is easy to cause sorting misjudgment; the capacity cycle test method: through complete charge and discharge test to measure the residual capacity and efficiency of the battery, is one of the more accurate sorting methods at present, but this method is time-consuming and requires high equipment, and is not suitable for large-scale rapid sorting. Model prediction method: based on equivalent circuit model, electrochemical model or data-driven model to model the battery state, and then sort according to the capacity, but the existing method also needs a long period of pre-test, and still cannot break through the low efficiency problem.

[0005] Although there are some technical solutions in the prior art for grouping batteries according to the state of charge of the batteries, they can only be applied to grouping of unretired power batteries and cannot be applied to large-scale gradient performance classification and grading. Some existing technologies use multiple charge and discharge and standing to sort the battery pack, but this sorting method takes a long time to test. Only using a single voltage static value has the defects of limited feature extraction dimension and inability to accurately reflect capacity differences, making it difficult to cope with the performance differentiation and complex aging state of the battery cell, and limiting the accuracy and generalization ability of capacity sorting. Some existing technologies disclose that the ohmic resistance and impedance characteristics of the retired battery are extracted through electrochemical impedance spectroscopy testing and relaxation time distribution analysis, the capacity is estimated by combining a black box model, and soft clustering sorting is realized by inputting multi-dimensional parameters into a Gaussian mixture model, which has certain rapid screening capability. However, this method has high requirements for testing equipment and operating conditions, and the electrochemical impedance spectroscopy measurement process is complex and time-consuming, making it difficult to meet the rapid screening needs of large quantities of retired batteries. Moreover, this relaxation time distribution analysis is an analytical method for impedance spectroscopy, and electrochemical impedance spectroscopy testing requires applying a low-frequency to high-frequency excitation signal to the battery, which requires professional testing equipment.

[0006] Therefore, there is a need for a rapid sorting method for gradient utilization batteries that can sort in large batches, has high sorting accuracy, sorts quickly, and is cost-effective. SUMMARY

[0007] The present application is to solve the defects of the existing retired battery sorting method, such as inaccurate sorting, low efficiency, high equipment requirements, and inability to sort in large batches, and provides a rapid sorting method for gradient utilization batteries that can sort in large batches, has high sorting accuracy, sorts quickly, and is cost-effective.

[0008] The rapid sorting method for gradient utilization batteries according to the present application comprises the following steps: S1, safety inspection of the gradient utilization batteries to be sorted; S2, charging the battery samples that pass the safety inspection to a preset cutoff voltage s under the same conditions and performing a standing test, and recording the standing voltage sequence during the standing process; S3, feature extraction of the collected standing voltage sequence to construct a sorting feature vector required for battery sorting; the sorting feature vector includes the energy, energy entropy, mean, and variance of the low-frequency approximate component; S4, dynamic modulation and gated weighted fusion of the feature vector F obtained by wavelet decomposition and the cutoff voltage s; S5, capacity testing of some battery samples to obtain their SOH values, and classification of the performance grade of the batteries according to the SOH using a gradient utilization battery sorting model; S6, training the cascade utilization battery sorting model based on the extracted modulation fusion features and performance levels, to obtain the trained cascade utilization battery sorting model; S7, obtaining the sorting feature vector F of the cascade utilization battery to be sorted and the preset cut-off voltage s, and inputting the trained battery sorting model to obtain the performance level prediction result of the battery.

[0009] Further, the sorting feature vector is the approximate component energy entropy based on discrete wavelet transform; the construction of the approximate component energy entropy based on discrete wavelet transform is as follows: S31, decomposing the static voltage sequence by using discrete wavelet transform to obtain a low-frequency approximate component; S32, calculating the energy entropy value.

[0010] Further, the energy of the low-frequency approximate component is calculated as follows: ; In the formula, E is the energy feature of the Lth layer low-frequency approximate component, is the Lth layer low-frequency approximate component obtained after wavelet decomposition, M is the sampling number, and i is the sampling point.

[0011] Further, the calculation formula of the energy entropy value is as follows: ; In the formula, H is the energy entropy value, E is the energy feature of the Lth layer low-frequency approximate component, M is the sampling number, and i is the sampling point.

[0012] Further, in S4, the mean value of the low-frequency approximate component is calculated as follows: ; In the formula, is the mean value of the low-frequency approximate component; is the Lth layer low-frequency approximate component obtained after wavelet decomposition, M is the sampling number, and i is the sampling point, i=1,2,...,M.

[0013] Further, in S4, the variance of the low-frequency approximate component is calculated as follows: ; In the formula, is the variance of the low-frequency approximate component; is the Lth layer low-frequency approximate component obtained after wavelet decomposition, M is the sampling number, and i is the sampling point, i=1,2,...,M.

[0014] Further, in S4, the specific process of dynamically modulating and gating and weighting fusion of the feature vector F obtained after wavelet decomposition and the cut-off voltage s is as follows: First, the extracted feature vector F and the cut-off voltage s are dynamically modulated: ; wherein, is the modulation feature, and is the modulation parameter determined by the cut-off voltage s, which is obtained by the following mapping function: ; wherein, and are weight matrices, and are bias vectors; The normalized attention weight vector is generated by the softmax gating function: ; wherein, is the weight matrix, is the bias vector, is the modulation feature, The function ensures that the weights satisfy ; Then, the feature weighting fusion is performed: ; wherein, z is the modulated and weighted fusion modulation fusion feature, is the i-th weight vector, is the i-th modulation feature, and N is the number of features.

[0015] Further, the performance level categories are divided according to the utilization scenarios.

[0016] The computer device disclosed by the application comprises a memory, a processor, a computer program stored on the memory and executable on the processor, and the processor executes the computer program to realize the fast sorting method of the battery for cascade utilization.

[0017] The computer readable storage medium disclosed by the application has a computer program stored thereon, and the computer program is executed by a processor to realize the fast sorting method of the battery for cascade utilization.

[0018] The beneficial effects of the application are: The application provides a fast sorting method for cascade utilization batteries. The method first performs discrete wavelet transform on a battery static voltage curve, realizes multi-scale decomposition, extracts energy, energy entropy and statistical characteristics and the like of a low-frequency component, and modulates and fuses different charging cutoff voltages and sorting characteristics. Then, the extracted characteristics are trained and sorted by combining a machine learning model, so that fast sorting of the battery SOH grade is realized. Compared with a traditional sorting method, the application has the following obvious advantages: first, the method can avoid complete charging and discharging test on large-scale retired batteries one by one, and greatly shortens the sorting period; second, in view of the actual problem that the initial voltages of the cascade utilization batteries are quite different, the application innovatively introduces a modulation and fusion mechanism of the cutoff voltage as a control means, and by adaptively adjusting the charging voltage interval of different batteries, the additional time consumption caused by charging all the batteries to the same cutoff voltage is avoided, so that the sorting efficiency and operation flexibility are significantly improved.

[0019] Specifically, by the cutoff voltage modulation technology, the optimal charging cutoff point can be determined according to the initial voltage state of each battery, unnecessary charging time is reduced, meanwhile, by the feature fusion algorithm, the voltage interval information and waveform characteristics are organically combined, the accuracy of the health state evaluation is enhanced. In addition, the application only needs to measure the voltage signal, the equipment investment cost is low, the implementation and deployment are convenient, and the application is particularly suitable for batch sorting and cascade utilization of large-scale retired batteries. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 A fast sorting flowchart of cascade utilization batteries; Figure 2 A curve diagram of static voltage sequences of two different cascade utilization batteries; Figure 3 A curve diagram of low-frequency approximate components obtained after discrete wavelet transform; Figure 4 A sorting result schematic diagram of the cascade utilization batteries. DETAILED DESCRIPTION

[0021] The following is only a preferred specific embodiment of the application, but the protection scope of the application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the application, which should be covered within the protection scope of the application. The following examples are only used to explain the application, and cannot be explained as a limitation on the application, the protection scope of the application should be subject to the protection scope of the claims. The embodiments of the application are described in detail below, in order to facilitate the description of the application and simplify the description, the technical terms used in the specification of the application should be interpreted broadly, including but not limited to the conventional replacement schemes not mentioned in the application, and including direct implementation and indirect implementation.

[0022] Embodiment 1 In combination Figures 1-4 To illustrate the present embodiment, the present embodiment discloses a method for rapid sorting of cascade utilization batteries, comprising the following steps: S1, first, safety inspection is performed on the cascade utilization batteries to be sorted, including whether there are obvious faults such as swelling, liquid leakage, short circuit, and tab deformation, and if the batteries with the abnormal conditions are found, they should be rejected and not enter the subsequent sorting process.

[0023] S2, the battery samples passing the safety inspection are charged to a preset cut-off voltage s under the same conditions and are subjected to a standing test. During the standing process, the sequence U(t) of the voltage change with time is recorded.

[0024] The same conditions include the same charging current rate, temperature, humidity, pressure, etc. The preset cut-off voltage s can be set to multiple different target voltages in actual application, such as 4.0V, 4.1V, 4.2V for ternary lithium batteries, and 3.4V, 3.5V, 3.6V for lithium iron phosphate batteries. According to the initial voltage state of the battery to be tested, the closest target cut-off voltage can be selected for charging, thereby reducing the charging time and improving the overall test efficiency.

[0025] Figure 2 is a standing voltage sequence graph drawn according to the sequences of the voltage change with time obtained by charging two different batteries to a cut-off voltage 2.8V and standing for 1 hour. The SOH of battery 1 is 0.768, and the SOH of battery 2 is 0.726. It can be seen that there is a significant difference in the curves of the standing voltage sequences of the two batteries: the voltage of battery 1 drops more gently, and the overall change amplitude is smaller, showing higher voltage stability, reflecting that the internal polarization effect is weaker and the recovery ability is stronger; while the voltage of battery 2 drops more significantly, indicating that its internal impedance is relatively large and the health status is poor. As can be seen, the voltage response characteristics in the standing process can effectively reflect the aging degree of the battery, providing an important basis for performance grade discrimination.

[0026] S3, the standing voltage sequence U(t) is subjected to feature extraction, and a feature vector required for battery sorting is constructed; the extracted sorting feature vector includes the energy, energy entropy, mean, and variance of the low-frequency approximation component based on discrete wavelet transform.

[0027] Specifically, the sorting feature vector is constructed as follows: First, the discrete wavelet transform is used to perform multi-scale decomposition on the standing voltage sequence, and the low-frequency approximation component is obtained.

[0028] Specifically, the Daubechies-6 wavelet function is used to perform L-layer discrete wavelet decomposition on the static voltage sequence U(t), to obtain an L-layer low-frequency approximation component , i is a sampling point.

[0029] S31, the energy of the low-frequency approximation component is calculated as follows: ; In the formula, E is the energy feature of the L-layer low-frequency approximation component, is the L-layer low-frequency approximation component obtained after wavelet decomposition, M is the number of samples, and i is a sampling point.

[0030] S32, the calculation formula of the energy entropy value is as follows: ; In the formula, H is the energy entropy value, E is the energy feature of the L-layer low-frequency approximation component, M is the number of samples, and i is a sampling point.

[0031] The mean of the low-frequency approximation component is calculated as follows: ; In the formula, is the mean of the low-frequency approximation component; is the L-layer low-frequency approximation component obtained after wavelet decomposition, M is the number of samples, i is a sampling point, and i=1, 2,..., M.

[0032] The variance of the low-frequency approximation component is calculated as follows: ; In the formula, is the variance of the low-frequency approximation component; is the L-layer low-frequency approximation component obtained after wavelet decomposition, M is the number of samples, i is a sampling point, and i=1, 2,..., M.

[0033] The wavelet transform has good time-frequency localization ability, which can effectively capture the non-stationary signal changes caused by polarization relaxation, potential drift and other reasons in the battery static process, so as to more accurately reflect the internal state and aging characteristics of the battery. Based on this, in the embodiment, the energy E, energy entropy H, mean and variance of the low-frequency approximation component are used as the quantitative indexes of the complexity of the energy distribution of the static voltage sequence in the low-frequency scale. The feature can depict the balance of the energy distribution in the approximation component, and is closely related to the polarization recovery behavior and health state of the battery.

[0034] In the embodiment, the number of layers L of the discrete wavelet decomposition is selected as 6. The number of layers is determined based on the sampling length of the curve of the battery static voltage sequence, the signal frequency range and the experimental results. The actual test shows that the low-frequency approximation component obtained by decomposing to the 6th layer can better retain the main trend characteristics of the voltage response, filter out high-frequency noise and random disturbance, and is conducive to the stability and accuracy of subsequent feature extraction and model identification.

[0035] It should be noted that the number of decomposition layers is not uniquely defined. Under different sampling frequencies or voltage platform conditions, the number of layers can also be adjusted according to the specific signal characteristics, for example, it can be selected as 5 layers or 7 layers, but generally 4-8 layers are appropriate to balance the feature expression ability and the computational complexity.

[0036] S4, dynamically modulating and gate-weighted fusing the sorted feature vector F obtained by wavelet decomposition and the preset cutoff voltage s.

[0037] First, the sorted feature vector F extracted and the preset cutoff voltage s are dynamically modulated: ; wherein, is the modulation feature, and are modulation parameters determined by the cutoff voltage s, which are obtained by the following mapping function: ; wherein, and are weight matrices, and are bias vectors. The transformation enables the model to adaptively adjust the feature distribution according to the cutoff voltage s, effectively reducing the distribution difference between data from different sources.

[0038] The normalized attention weight vector is generated by the softmax gate function: ; wherein, is the weight matrix, is the bias vector, is the modulation feature, The function ensures that the weight satisfies .

[0039] Then, the feature weighted fusion is performed: ; wherein, z is the modulation fusion feature after modulation and weighted fusion, is the i-th weight vector, is the i-th modulation feature, and N is the number of features.

[0040] The modulation mechanism can adaptively correct and align the resting feature vector according to the initial voltage of the battery and the charging cutoff condition. The low-frequency approximate component of the resting voltage sequence reflects the internal polarization effect and capacity degradation characteristics of the battery, while different initial voltages and charging conditions can cause shifts in feature amplitude and distribution. By applying modulation and attention weight fusion to the feature vector, the model can effectively reduce the feature differences under different test conditions, maintain the consistency and stability of feature expression, and thus improve the universality and robustness of the sorting model in cross-batch batteries and diversified application scenarios.

[0041] Secondly, in the model application stage, the battery does not need to be strictly charged to a fixed cutoff voltage, but only needs to be charged to any voltage within the preset interval to complete feature acquisition. This strategy significantly reduces the charging time consumption, improves the overall sorting efficiency, and at the same time ensures the sorting accuracy, avoiding the constraints of harsh test conditions on large-scale retired battery batch sorting.

[0042] S5, randomly select part of the battery samples from the batteries that pass the safety check for capacity test, obtain the SOH value, and classify the performance level of the battery according to the SOH.

[0043] The SOH value is the ratio of the test capacity to the rated capacity, which is used to represent the health degree of the battery. The performance level category can be divided according to the scene of ladder utilization, for example, SOH belongs to [0.8, 0.85) is class one, belongs to [0.75-0.8) is class two, and belongs to [0.7-0.75) is class three. A large number of literature and industry standards (such as T / CASE 240-2021) take SOH≈0.8 as the typical inflection point for the battery to enter the "ladder utilization stage" from the "power battery stage". Due to the inconsistency between batteries, there is a large difference in the capacity of retired batteries. The interval division principle is as follows: SOH belongs to [0.8, 0.85): The capacity of the battery in this range decays less, and the output power and cycle stability are still good. After consistency screening and grouping, it can meet the demand of grid-side energy storage for service life and safety. SOH belongs to [0.75, 0.8): The capacity has decreased, but it can still guarantee a long standby power supply. The base station and other scenes mainly run in the intermittent mode of "standby-discharge-charge", so the continuous power requirement is not high, and therefore the batteries in this interval are suitable for use as standby power sources in base stations. SOH belongs to [0.7, 0.75): The capacity of the battery in this interval further decays, and the cycle life and consistency are insufficient to support high-power or long-time applications, but it can still meet the demand of low-speed electric vehicles or light energy storage devices. When SOH<0.7, the battery performance degrades obviously, the internal resistance increases, and the safety risk increases, so it is not suitable as a ladder utilization battery and generally enters the material disassembly and recycling link. The above description is for a typical application scenario, and in practice, it can be set according to the specific scene.

[0044] S6, constructing and pre-training the cascade utilization battery sorting model based on the obtained modulation fusion features and SOH performance classification.

[0045] Optionally, the cascade utilization battery sorting model is constructed based on a machine learning method. In this embodiment, the battery sorting model is trained by random forest. The sorting model constructed based on the random forest model is trained, and the training data includes the modulation and weighted fusion fusion features z obtained from steps S1-S4 and the performance level classification label. The classification label can be set according to the capacity distribution of the cascade utilization battery and the specific use scene. The fusion features z and the performance level classification label are input into the random forest model to train the sorting model, and a cascade utilization battery sorting model based on the relaxation voltage fusion features is obtained.

[0046] As an ensemble learning method, random forest constructs multiple decision trees with differences, and in the prediction stage, the judgment results of each sub-model are voted and fused to realize the classification output of the overall model, thereby improving the stability and accuracy of the overall model. This method is not sensitive to noise, has strong nonlinear modeling ability and good anti-overfitting performance, and can effectively process complex data scenes with high feature dimension and uneven sample distribution.

[0047] In the classification model of the random forest algorithm, a commonly used way is to directly count the prediction results of multiple decision trees and determine the final output classification label by majority voting. That is, each decision tree independently judges the battery performance level, and the final result is determined by the voting results of all decision trees. This mechanism can form a complement between different decision trees, effectively alleviate the deviation that may be caused by a single model, and thereby significantly improve the accuracy and robustness of the sorting result.

[0048] Specifically, each decision tree gives a prediction classification label for the input sample, and the final prediction result of the random forest model is determined by the following formula: ; wherein, is the final sorting category, B is the total number of decision trees, the function is to return the category with the highest frequency in all prediction results, is the category of the first decision tree, is the category of the second decision tree, is the category of the Bth decision tree.

[0049] S7, obtaining the feature vector and the cutoff voltage of the cascade utilization battery to be sorted, and inputting the trained battery sorting model to obtain the performance level prediction result of the battery, thereby realizing rapid sorting.

[0050] After the sorting model training in step S6 is completed, only the feature vector of the to-be-sorted cascade utilization and the corresponding cut-off voltage are extracted through steps S2 and S3 and input into the trained sorting model, and the SOH performance class of the battery sample can be obtained.

[0051] As Figure 3 The verification result diagram of the fast sorting of the cascade utilization battery is shown. Steps S1-S6 are performed on 130 cascade utilization batteries. In order to verify the sorting effect, capacity tests are performed on all the batteries to determine the real SOH values. In this embodiment, the batteries are divided into three classes according to SOH, SOH belonging to [0.8, 0.85) is class 1, SOH belonging to [0.75-0.8) is class 2, and SOH belonging to [0.7-0.75) is class 3. 104 batteries are randomly selected from the 130 cascade utilization batteries as training samples. According to step S2, the cut-off voltage is set to 3.4V and 3.6V, and the batteries are charged to the nearest cut-off voltage and left for 1 hour, wherein the sampling frequency of the voltage is 1Hz. According to steps S3 and S4, the features are extracted and modulated and fused. The modulated and fused features of the 104 batteries and the corresponding SOH performance classification are trained using a random forest. According to steps S2 and S3, the remaining 26 batteries are charged to any voltage value in the set 3.4V-3.6V cut-off voltage interval according to the initial voltage state of the battery, and the feature vector and the corresponding cut-off voltage are obtained. According to step S6, the 26 test battery samples are input into the sorting model trained by the 104 batteries, and the SOH level of the 26 test battery samples is obtained. As shown in Figure 4 It can be seen that the model sorting result is highly consistent with the actual label, and only a few boundary samples deviate, the overall sorting accuracy is high, which verifies the effectiveness and reliability of the method for sorting the cascade utilization batteries. In practical application, the initial voltage of the to-be-sorted battery can be selected to charge the closest preset cut-off voltage interval, and the sorting feature vector input and the cut-off voltage are input into the trained model to realize the fast sorting of the cascade utilization battery performance level.

[0052] Embodiment 2 A computer device, comprising: a memory, a processor to store a computer program on the memory and executable on the processor, and the processor executes the computer program to implement the fast sorting method of the cascade utilization battery.

[0053] Embodiment 3 A computer readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the fast sorting method of the cascade utilization battery.

Claims

1. A rapid sorting method of cascade utilization battery, characterized in that, The method comprises the following steps: S1, performing safety inspection on the to-be-sorted cascade utilization battery; S2, charging the battery sample that passes the safety inspection to a preset cut-off voltage s under the same condition and performing a standing test, and recording a standing voltage sequence during the standing process; S3, performing feature extraction on the standing voltage sequence to construct a sorting feature vector required for battery sorting; the sorting feature vector comprises energy, energy entropy, mean value, and variance of the low-frequency approximation component; S4, performing dynamic modulation and gated weighted fusion on the sorting feature vector F obtained through wavelet decomposition and the preset cut-off voltage s; S5, selecting part of the battery sample to perform capacity test to obtain an SOH value, and classifying the performance grade of the battery according to the SOH by using the cascade utilization battery sorting model; S6, training the cascade utilization battery sorting model based on the extracted modulation fusion feature and the performance grade to obtain the trained cascade utilization battery sorting model; S7, obtaining the sorting feature vector F of the to-be-sorted cascade utilization battery and the preset cut-off voltage s, and inputting the trained battery sorting model to obtain the performance grade prediction result of the battery.

2. The method of claim 1, wherein the method is characterized by: The sorting feature vector is an energy entropy of an approximation component based on discrete wavelet transform; the energy entropy of the approximation component based on the discrete wavelet transform is constructed as follows: S31, decomposing the standing voltage sequence by using the discrete wavelet transform to obtain a low-frequency approximation component; S32, calculating an energy entropy value.

3. The method of claim 2, wherein the step of determining the state of charge of each battery is performed by measuring the voltage of each battery. The energy of the low-frequency approximation component is calculated as follows: ； In the formula, E is an energy feature of the Lth layer low-frequency approximation component, is the Lth layer low-frequency approximation component obtained after wavelet decomposition, M is the number of samples, and i is a sampling point.

4. The method of claim 2, wherein the method is characterized by: The calculation formula of the energy entropy value is as follows: ； In the formula, H is the energy entropy value, E is the energy feature of the Lth layer low-frequency approximation component, M is the sampling number, and i is a sampling point.

5. The method of claim 1, wherein the method is characterized by: In S4, the mean value of the low-frequency approximation component is calculated as follows: ； wherein is the mean of the low frequency approximation component; is the Lth level low frequency approximation component after wavelet decomposition, M is the number of samples, i is the sampling point, i = 1, 2,..., M.

6. The method of claim 1, wherein the method is characterized by: In S4, the variance of the low-frequency approximation component is calculated as follows: ； wherein is the low frequency approximation component variance; is the Lth low frequency approximation component after wavelet decomposition, M is the number of samples, i is the sampling point, i = 1, 2,..., M.

7. The method of claim 1, wherein the method is characterized by: In S4, the specific process of performing dynamic modulation and gated weighted fusion on the sorting feature vector F obtained through wavelet decomposition and the cut-off voltage s is as follows: First, the extracted feature vector F and the cut-off voltage s are dynamically modulated: ； wherein is a modulation feature, and are modulation parameters determined by the cut-off voltage s, respectively, which are obtained from the following mapping function: ； wherein, and are weight matrices, and are bias vectors, respectively. A normalized attention weight vector is generated by using a softmax gating function: ； wherein is a weight matrix, is a bias vector, is a modulation feature, the function ensures that the weights satisfy ; Then, feature weighted fusion is performed: ； In the formula, z is a modulation fusion feature after modulation and weighted fusion, is the i th weight vector, is the i th modulation feature, and N is the number of features.

8. The method of claim 1, wherein the method is a method of rapid sorting of batteries for use in a cascade. The category of the performance grade is divided according to the scene of cascade utilization.

9. A computer apparatus, comprising: It comprises: a memory and a processor to store a computer program on the memory and run the computer program on the processor, and the processor executes the computer program to implement the fast sorting method of the cascade utilization battery according to any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer program is stored on the memory and is executed by the processor to implement the fast sorting method of the cascade utilization battery according to any one of claims 1-8.

Images