Retired power battery sorting method considering battery aging state distribution imbalance

Abstract

The invention discloses a retired power battery sorting method considering battery aging state distribution imbalance. The method comprises the following steps: 1, measuring and recording end voltage of a retired battery by using voltage measuring equipment; 2, randomly extracting a part of data of the current batch of retired power batteries as a battery for testing; 3, performing a test experiment on the tested battery based on the design step, and collecting and storing data; 4, performing data cleaning and preprocessing on the test data, and then executing designed adaptive feature engineering; and 5, pre-training the complementary energy estimation model embedded with the feature distribution smoothing module by using the output in the previous step. And 6, repeating the steps 3-4 on the remaining batteries, and deploying a pre-training model to execute complementary energy estimation. According to the method, errors caused by unbalanced aging distribution can be overcome, so that the time cost and the energy consumption cost of retired battery sorting work are reduced on the premise of ensuring the sorting accuracy.

Description

Technical Field

[0001] The present invention belongs to the field of cascade utilization of retired batteries, and specifically is a retired power battery sorting method taking into account the unbalanced distribution of battery aging states. Background Art

[0002] Power batteries are important energy storage components for new energy vehicles. They are usually composed of a large number of single cells connected in series and parallel to form modules, which are then connected in series to form a battery pack to provide power for the vehicle. After a long period of on-board use and then retired, most of the retired power batteries are disassembled and enter the retirement process because the available capacity drops to the threshold of 80% of the original capacity; and there is also a considerable number of battery types composed of over-aged power batteries and healthy batteries that are retired due to large inconsistencies with other batteries in the on-board battery pack. This type of aging state imbalance will directly affect the accuracy of the subsequent residual energy estimation model and cause sorting errors and inconveniences in cascade utilization. Therefore, it is necessary to accurately estimate the residual energy of these batteries with different degrees of aging before the retired power batteries enter the cascade utilization environment, so as to achieve accurate sorting.

[0003] At present, the steps for sorting retired batteries are roughly as follows: 1) Manually conduct a preliminary screening of the structure and appearance of retired batteries, and screen out batteries that are obviously damaged and deformed and cannot be used in cascades; 2) Use test equipment to measure the performance parameters of retired batteries, and establish corresponding databases for different models and types of batteries to store their test data. At the same time, batteries that are severely aged and cannot meet the requirements of any subsequent cascade utilization scenarios (such as SOH below 40%) are included in the disassembly and recycling process; 3) Sort the batteries in each database according to specific indicator parameters, and classify batteries with similar performance parameters into one category to ensure the consistency of subsequent grouping work.

[0004] Most of the current retired power battery sorting methods require the recycled retired power batteries to be discharged once and then fully charged and fully discharged. They are classified according to the available capacity and other performance parameters obtained from the test, and then classified and stored in the corresponding categories, and then supplied according to the needs of different tiered utilization scenarios. However, the existing sorting methods all have a common blind spot, which is the lack of consideration for the imbalanced distribution of battery aging status within the same batch. This type of imbalance problem will lead to an imbalance in the test data and affect the effectiveness of the sorting method, thereby reducing the performance of retired batteries in tiered utilization scenarios. The use of data-driven methods to sort retired batteries needs to consider the impact of such imbalances, otherwise the model will be affected by the knowledge of the battery characteristics of the majority class aging state during learning, resulting in the trained model having too large an estimation error for the minority class model. Summary of the invention

[0005] The present invention aims to solve the blind spots and shortcomings in the current retired battery sorting technology summarized above. By proposing a retired power battery sorting method that takes into account the unbalanced distribution of battery aging status, it is hoped that the influence of the imbalance phenomenon that is prevalent in such actual retired battery sorting industrial scenarios on the sorting results can be overcome, and full charge and discharge testing of large quantities of retired batteries can be avoided and redundant testing processes can be reduced, thereby reducing the energy consumption and time cost of retired battery sorting.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme:

[0007] The method for sorting retired power batteries in consideration of the imbalanced distribution of battery aging states of the present invention is characterized in that it comprises the following steps:

[0008] Step 1: Perform appearance screening on a batch of retired power batteries to obtain the remaining retired batteries after screening;

[0009] Step 2: Randomly select the remaining retired batteries and obtain a total of Test battery of

[0010] Step 3: Collect test data of the test battery;

[0011] Step 3.1: After the test battery is left to stand for a period of time, the initial terminal voltage of the test battery is measured. ,in, Representative The terminal voltage of a test battery;

[0012] Step 3.2: Construct a charging and discharging strategy and use it to Test batteries to get the battery test data set ,in, Indicates Dynamic characteristic data of the test battery, Indicates Static characteristic data of the test battery;

[0013] Step 4: After preprocessing, the standardized feature sequence is obtained ;in, Represents the The standardized feature sequences of the test batteries;

[0014] Step 5: Input into LASSO model for processing, and get the A preliminary characterization sequence of test batteries;

[0015] Step 6: After any two features in the preliminary feature sequence of the test battery are randomly combined, grey correlation analysis is performed to select the features with high correlation coefficient to form the first Optimal feature sequence for a test battery ,in, Representative The kth best feature of the test battery, represents the number of optimal features;

[0016] Step 7: Build a BiLSTM-based time series regression prediction model, including: feature extraction module, feature smoothing module, residual energy estimation module and temperature-based correction module, and use it to obtain the optimal feature sequence of n test batteries. Processing is performed to obtain the estimated remaining energy of the retired battery ,in, represents the estimated value of the remaining energy of the mth test battery;

[0017] Step 8: Use formula (8) to obtain the corrected residual energy sequence :

[0018] (8)

[0019] In formula (8), is the influence coefficient of temperature on capacity; is the reference temperature; To test the ambient temperature; represents the corrected residual energy value of the mth test battery;

[0020] Step 9: Construct a comprehensive loss function according to formula (9) :

[0021] (9)

[0022] In formula (9), are 2 balance coefficients, represents the loss function of the residual energy estimation, Represents the loss function of the feature smoothing process;

[0023] Step 10: Use the Adam optimizer to train the time series regression prediction model and calculate the comprehensive loss function To update the model parameters until Until convergence, the optimal retired battery remaining energy estimation model is obtained, which is used to estimate the remaining energy of the retired batteries, so as to estimate the remaining energy of the battery according to the obtained battery remaining energy and the calculated average dynamic internal resistance. Realize battery sorting.

[0024] The method for sorting retired power batteries taking into account the imbalanced distribution of battery aging states described in the present invention is also characterized in that step 3.2 comprises the following steps:

[0025] Step 3.2.1: The low current mixed pulse power characteristics test is performed on the test battery to obtain dynamic characteristic data that can characterize the dynamic characteristics of the battery. ;in, represents the pulse current, and , represents the pulse current of the i-th rate, represents the voltage change during the pulse test, and , It represents the voltage change of the mth test battery in the i-th pulse current, represents the recovery voltage, and , represents the recovery voltage of the mth test cell, represents the time to reach a stable recovery voltage, and , It indicates the time when the mth test battery reaches the stable recovery voltage. Indicates the final voltage at which the test is completed, and ; represents the terminal voltage of the mth test battery;

[0026] Step 3.2.2 Perform partial charging test on the test battery to obtain static charging data that can characterize the battery aging status information ;in, Indicates the initial state of charge after discharge. Represents the set charging termination state of charge; Indicates the voltage data during the charging process, and , Indicates the charging voltage data of the mth test battery; Indicates the charging capacity data during the charging process, and , Represents the charging capacity data of the mth test battery; Indicates the charging energy data during the charging process, and , Represents the charging energy data of the mth test battery; represents the terminal voltage after partial charging is completed, and , represents the terminal voltage of the mth test battery;

[0027] 3. According to claim 2, a method for sorting retired power batteries taking into account the imbalanced distribution of battery aging states, characterized in that step 3.2.1 is to sort the first After the test battery is charged to the initial state charge with a constant current, it is left to stand for a period of time; then the set pulse current is After standing Perform pulse discharge test on each test battery and record the Voltage changes of the test cells during the pulse process ,in, represents the pulse current of the ith rate, Indicates the pulse current of the i-th rate Next The voltage change of the test battery, z represents the number of rates;

[0028] After the pulse process is completed, The test battery is left to stand until the battery voltage stabilizes, and the corresponding recovery voltage is recorded. , restore the rest time and open circuit voltage .

[0029] Further, step 3.2.2 is to The test cells were discharged to their initial state of charge , and record the initial voltage at this time , according to the set discharge current ratio After the first test battery is discharged, The test cell is charged at a constant current Charge to the set state of charge , and record the first Test battery voltage , Charging capacity and charging energy After charging is completed, Let the test battery stand for a while and record the Final open circuit voltage of a test cell after partial charging .

[0030] Furthermore, the step 4 includes:

[0031] Step 4.1: After data cleaning, we get a clean data set ;

[0032] Step 4.2: Based on , build a basic feature library ,in, Indicates The dynamic characteristics of the test battery, Indicates A static characterization set of test batteries;

[0033] Step 4.3: The feature sequence composed of all dynamic features and static features of each test battery and the capacity label of the test battery After splicing and normalization preprocessing, the standardized feature sequence is obtained ,in, represents the capacity label of the mth tested battery;

[0034] Furthermore, step 4.2 includes the following steps:

[0035] Step 4.2.1: The low current mixed pulse power characteristic test data in the image is used for feature extraction to obtain a dynamic feature set. ;

[0036] Step 4.2.2: The voltage, energy, and capacity data obtained from the partial charging test in the test are averaged, the maximum and minimum extreme values ​​are taken, and used to calculate the variance, kurtosis, skewness, quantile, and slope of the curve to obtain a static feature set. ,in, Indicates The voltage-related characteristics of the test cells, Indicates Capacity-related characteristics of the test cells, Indicates Energy-related characteristics of the test cells, Indicates Partially charged IC curve related characteristics of the tested cells.

[0037] Furthermore, step 4.2.1 includes the following steps:

[0038] Step 4.2.1.1: According to Voltage change of the test cell With pulse current Dynamic internal resistance obtained by doing business , using formula (1) and formula (2) to obtain The average dynamic internal resistance of the tested batteries and dynamic internal resistance change rate and accordingly as the First dynamic characterization of the test battery and the second dynamic feature :

[0039] (1)

[0040] (2)

[0041] In formula (1) and formula (2), represents the dynamic internal resistance calculated under the i-th pulse current, and , represents the dynamic internal resistance of the mth test battery under the current pulse current, Indicates the maximum internal resistance, Indicates the minimum value of internal resistance;

[0042] Step 4.2.1.2: Based on the recovery voltage , recovery time and the initial terminal voltage , using formula (3) and formula (4) to obtain The recovery voltage difference of the tested batteries and pulse recovery rate and accordingly as the The third dynamic characteristics of the test battery and the fourth dynamic feature , thus obtaining the dynamic feature set of n test batteries ;

[0043] (3)

[0044] (4)

[0045] In formula (3) and formula (4), represents the recovery voltage of the mth test battery; represents the recovery time of the mth test battery; Represents the initial terminal voltage of the mth test battery.

[0046] Furthermore, step seven includes the following steps:

[0047] Step 7.1: The feature extraction module uses a BiLSTM layer to Process and output Deep features of the test battery ,in, Indicates Test battery No. deep features, Represents the total number of deep features;

[0048] Step 7.2: The feature smoothing module is based on the feature distribution smoothing strategy Processing is performed to obtain a smoothed feature sequence ;

[0049] Step 7.3: The residual energy estimation module is composed of a fully connected layer and a regression estimation layer; wherein a fully connected layer is Perform feature mapping to generate the The regression estimation layer is based on the regression function of the mapping features of the test battery The mapping features of the test battery are processed to generate the Estimated remaining energy of the test battery .

[0050] Furthermore, step 7.2 includes the following steps:

[0051] Step 7.2.1: Use equations (5) and (6) to get the updated feature mean and the updated feature variance :

[0052] (5)

[0053] (6)

[0054] In formula (5) and formula (6), represents the smoothing coefficient; and Respectively represent The feature means and feature variances of the test batteries before updating;

[0055] Step 7.2.2: Use equation (7) to Smoothing is performed to obtain the smoothed feature sequence ;

[0056] (7)

[0057] In formula (7), To prevent a tiny constant with a zero denominator; Indicates The first test battery after smoothing Deep features;

[0058] Furthermore, in step 9, the loss function of the residual energy estimation is obtained by using equations (10) and (11) respectively: And the loss function of the feature smoothing process ;

[0059] (10)

[0060] (11).

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] 1. The scheme of the present invention collects test data by designing a test step specifically for testing the dynamic characteristics of the battery and a local charge and discharge experiment. The remaining energy estimation and subsequent sorting work are performed only through local data, thereby avoiding the first step of emptying and full charge and discharge testing of retired batteries, significantly eliminating redundant test steps in the sorting process, and reducing the test energy consumption and sorting time of retired battery manufacturers.

[0063] 2. The present invention designs an adaptive feature engineering, which includes all dynamic and static feature information based on local test data, and uses eight types of statistical processing and IC curve-related features to jointly construct a complete feature library and optimization pipeline containing battery dynamic performance and aging information, and automatically screen out the optimal feature combination for different types of batteries, thereby improving the generalization ability of the estimation model and the adaptability to various retirement sorting scenarios for the subsequent remaining energy estimation of retired batteries.

[0064] 3. The residual energy estimation model in the method of the present invention is based on a BiLSTM neural network, and is embedded with an FDS layer (Feature Distribution Smooth) for smoothing feature distribution during the training of the neural network model to overcome the impact of imbalance, thereby improving the residual energy estimation accuracy and the residual energy-based sorting accuracy after the subsequent pre-trained residual energy estimation model is deployed.

[0065] 4. The test process of the present invention uses the charge and discharge data obtained from the low-current hybrid pulse power characteristic (L-HPPC) test and the partial charging test for feature extraction and model training as well as subsequent residual energy estimation, which not only ensures the uniformity of the test environment and data attributes of each test battery, but also avoids the redundant test steps caused by fully charging and discharging all batteries.

[0066] 5. The method adopted in the present invention is to assist in sorting based on two indicators: residual energy estimation by data-driven method and average dynamic internal resistance calculated from dynamic characteristics. Compared with the current method of obtaining residual energy by full charge and full discharge, this solution is more suitable for use in factory testing processes, reducing the complexity of manual screening work, and significantly improving sorting accuracy and efficiency. It is universal for sorting scenarios of various types of batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 This is a sorting structure diagram of the equipment involved in the present invention;

[0068] Figure 2 It is a flow chart of the whole sorting method of the present invention;

[0069] Figure 3 A flow chart of residual energy estimation used in the present invention;

[0070] Figure 4 Schematic diagram of the flow of the feature distribution smoothing method. DETAILED DESCRIPTION

[0071] In this embodiment, a sorting system structure is used in a method for sorting retired power batteries taking into account the imbalanced distribution of battery aging states, such as Figure 1 As shown, it includes: voltage measurement equipment, charge and discharge test cabinet, data acquisition equipment, and storage and computing equipment. The functions of each device are as follows:

[0072] The voltage measurement equipment is used to quickly measure the current voltage value of retired batteries with different residual energy, without the need to charge and discharge to obtain the curve;

[0073] The charge and discharge test cabinet is the most commonly used test equipment for a large number of single cells. It has the characteristics of multi-channel and high power. It can send corresponding steps to each channel battery to perform charge and discharge tests through the host computer settings.

[0074] The data acquisition device is responsible for collecting battery test data, such as current, voltage, energy, etc., and external physical parameters, such as test temperature, battery size change, etc.;

[0075] The data storage and computing devices are generally integrated in the same host computer. The storage device stores the collected battery information, and the computing device is responsible for data processing and the training and application of the remaining energy estimation model. The flow chart of the entire sorting method is shown in the figure. Figure 2 As shown, the specific steps include:

[0076] Step 1: Perform appearance screening on a batch of retired power batteries to obtain the remaining retired batteries after screening; the current state of charge of the retired batteries is not considered, and there is no need to discharge the remaining battery energy.

[0077] Step 2: Test batteries are randomly selected, and the total amount obtained by random sampling is The retired batteries are used as test batteries.

[0078] Step 3: Collect test data of the test battery;

[0079] Step 3.1: After the test battery is left to stand for a period of time, the initial terminal voltage of the test battery is measured. ,in, Representative The terminal voltage of a test battery;

[0080] Step 3.2: Construct a charging and discharging strategy and use it to Test batteries to get the battery test data set ,in, Indicates Dynamic characteristic data of the test battery, Indicates Static characteristic data of the test battery;

[0081] Step 3.2.1: The low current mixed pulse power characteristics test is performed on the test battery to obtain dynamic characteristic data that can characterize the dynamic characteristics of the battery. ;in, represents the pulse current, and , represents the pulse current of the i-th rate, represents the voltage change during the pulse test, and , It represents the voltage change of the mth test battery in the i-th pulse current, represents the recovery voltage, and , represents the recovery voltage of the mth test cell, represents the time to reach a stable recovery voltage, and , It indicates the time when the mth test battery reaches the stable recovery voltage. Indicates the final voltage at which the test is completed, and ; Represents the terminal voltage of the mth test battery.

[0082] No. After the test battery is charged to the initial state charge with a constant current, it is left to stand for a period of time; then the set pulse current is After standing Perform pulse discharge test on each test battery and record the Voltage changes of the test cells during the pulse process ,in, represents the pulse current of the ith rate, Indicates the pulse current of the i-th rate Next The voltage change of the test battery, z represents the number of rates;

[0083] After the pulse process is completed, The test battery is left to stand until the battery voltage stabilizes, and the corresponding recovery voltage is recorded. , restore the rest time and open circuit voltage .

[0084] Step 3.2.2 Perform partial charging test on the test battery to obtain static charging data that can characterize the battery aging status information ;in, Indicates the initial state of charge after discharge. Represents the set charging termination state of charge; Indicates the voltage data during the charging process, and , Indicates the charging voltage data of the mth test battery; Indicates the charging capacity data during the charging process, and , Represents the charging capacity data of the mth test battery; Indicates the charging energy data during the charging process, and , Represents the charging energy data of the mth test battery; represents the terminal voltage after partial charging is completed, and , Represents the terminal voltage of the mth test battery.

[0085] The first The test cells were discharged to their initial state of charge , and record the initial voltage at this time , according to the set discharge current ratio After the first test battery is discharged, The test cell is charged at a constant current Charge to the set state of charge , and record the first Test battery voltage , Charging capacity and charging energy After charging is completed, Let the test battery stand for a while and record the Final open circuit voltage of a test cell after partial charging .

[0086] Step 3.3: Use data acquisition equipment to collect the charging and discharging data of retired batteries in each channel during the test and the ambient temperature data during the test, and transmit them to the host computer through protocol communication for storage and backup.

[0087] Step 4: Data preprocessing and feature engineering;

[0088] Step 4.1: After data cleaning, we get a clean data set ;

[0089] Step 4.2: Based on , build a basic feature library ,in, Indicates The dynamic characteristics of the test battery, Indicates A static characterization set of test batteries;

[0090] Step 4.2.1: The low current mixed pulse power characteristic test data in the image is used for feature extraction to obtain a dynamic feature set. ;

[0091] Step 4.2.1.1: According to Voltage change of the test cell With pulse current Dynamic internal resistance obtained by doing business , using formula (1) and formula (2) to obtain The average dynamic internal resistance of the tested batteries and dynamic internal resistance change rate and accordingly as the First dynamic characterization of the test battery and the second dynamic feature :

[0092] (1)

[0093] (2)

[0094] In formula (1) and formula (2), represents the dynamic internal resistance calculated under the i-th pulse current, and , represents the dynamic internal resistance of the mth test battery under the current pulse current, Indicates the maximum internal resistance, Indicates the minimum internal resistance.

[0095] Step 4.2.1.2: Based on the recovery voltage , recovery time and the initial terminal voltage , using formula (3) and formula (4) to obtain The recovery voltage difference of the tested batteries and pulse recovery rate and accordingly as the The third dynamic characteristics of the test battery and the fourth dynamic feature , thus obtaining the dynamic feature set of n test batteries ;

[0096] (3)

[0097] (4)

[0098] In formula (3) and formula (4), represents the recovery voltage of the mth test battery; represents the recovery time of the mth test battery; Represents the initial terminal voltage of the mth test battery.

[0099] Step 4.2.2: The voltage, energy, and capacity data obtained from the partial charging test in the test are averaged, the maximum and minimum extreme values ​​are taken, and used to calculate the variance, kurtosis, skewness, quantile, and slope of the curve to obtain a static feature set. ,in, Indicates The voltage-related characteristics of the test cells, Indicates Capacity-related characteristics of the test cells, Indicates Energy-related characteristics of the test cells, Indicates The relevant characteristics of the partial charge IC curve of the test battery;

[0100] Use the above feature sequences to form a static feature library The specific static statistical feature calculation method can be implemented by formulas (5)-(12):

[0101] Maximum value:

[0102] (5)

[0103] Minimum:

[0104] (6)

[0105] Mean:

[0106] (7)

[0107] Kurtosis:

[0108] (8)

[0109] Skewness:

[0110] (9)

[0111] Quantile:

[0112] (10)

[0113] variance:

[0114] (11)

[0115] Slope:

[0116] (12)

[0117] In the above statistical calculation formula, is the data sequence of the dth sample; The number of data points representing the sample; represents the quantile, represents the lower bound of the quantile, is the frequency is the group spacing.

[0118] The IC curve correlation calculation and feature selection are shown in formula (13)-(14):

[0119] The IC curve is the ratio of the change in capacity to the change in terminal voltage. is the total coordinate, the corresponding voltage The horizontal axis is the curve of the internal electrochemical reaction of the reaction battery aging, where:

[0120] (13)

[0121] In formula (13), and Respectively indicate the corresponding Battery capacity and terminal voltage at the moment; In its discrete form. In its constant current charge and discharge stage, the calculation formula can be further converted into:

[0122] (14)

[0123] In formula (14), is the current in the constant current stage, is the sampling interval, is the sampling time.

[0124] According to the calculated IC curve, three indicators including the peak value on the IC curve and its corresponding position, and the area under the IC curve are selected as relevant features of the IC curve.

[0125] Step 4.3: The feature sequence composed of all dynamic features and static features of each test battery and the capacity label of the test battery To splice, and , Represents the capacity label of the mth test battery; after normalization preprocessing, the standardized feature sequence is obtained ,in, Represents the The standardized feature sequence of the test battery; The feature sequence composed of all dynamic features and static features of each test battery is concatenated with the capacity label of the test battery, and then normalized and preprocessed to obtain the standardized feature sequence. ,in, Represents the The standardized feature sequences of the test batteries;

[0126] Will Input into LASSO model for processing, and get the A preliminary characterization sequence of test batteries;

[0127] For After any two features in the preliminary feature sequence of the test battery are randomly combined, grey correlation analysis is performed to select the features with high correlation coefficient to form the first Optimal feature sequence for a test battery ,in, Representative The kth best feature of the test battery, Represents the number of optimal features.

[0128] The optimization algorithm used is the least absolute shrinkage and selection operator (LASSO). Based on the commonly used multivariate linear regression, LASSO compresses the coefficients of the features by adding penalty coefficients, adjusts the regression coefficients of low-correlation features to zero, and then achieves feature selection. The results of LASSO feature optimization are then subjected to grey relational analysis (GRA). The feature combination with higher correlation is selected as the final input for residual energy estimation.

[0129] Step 5: Build a BiLSTM-based time series regression prediction model, including: feature extraction module, feature smoothing module, residual energy estimation module and temperature-based correction module, and use it to obtain the optimal feature sequence of n test batteries. Processing is performed to obtain the estimated remaining energy of the retired battery ,and , Represents the estimated remaining energy of the mth test battery.

[0130] Step 5.1: The feature extraction module uses a BiLSTM layer to Process and output Deep features of the test battery ,in, Indicates Test battery No. deep features, Represents the total number of deep features; Figure 3 As shown;

[0131] Step 5.2: If Figure 4 As shown in the figure, the feature smoothing module is based on the feature distribution smoothing strategy. Processing is performed to obtain the smoothed feature sequence :

[0132] Step 5.2.1: Use equations (15) and (16) to get the updated feature mean and the updated feature variance :

[0133] (15)

[0134] (16)

[0135] In formula (15) and formula (16), represents the smoothing coefficient; represents the mth test battery Features. They represent the feature mean and feature variance of the mth test battery before updating.

[0136] Step 5.2.2: Use equation (17) to Each feature in the deep feature sequence Smoothing is performed to obtain the smoothed feature sequence ;

[0137] (17)

[0138] In formula (17), To prevent a tiny constant with a zero denominator, Indicates The first test battery after smoothing Deep features;

[0139] Step 5.3: The residual energy estimation module consists of a fully connected layer and a regression estimation layer; Perform feature mapping to generate the The regression estimation layer maps the features of the test battery based on the regression function The mapping features of the test battery are processed to generate the Estimated remaining energy of the test battery ;

[0140] Step 5.4: Use formula (18) to obtain the corrected residual energy sequence ,and , Represents the corrected residual energy value of the mth test battery:

[0141] (18)

[0142] In formula (18), is the coefficient of influence of temperature on capacity, which can be taken as 0.0075 in this example ; is the reference temperature; To test the ambient temperature.

[0143] Step 5.5: Construct a comprehensive loss function according to formula (19) :

[0144] (19)

[0145] In formula (19), are 2 balance coefficients, represents the loss function of the residual energy estimation, represents the loss function in the feature smoothing process, and has:

[0146] (20)

[0147] (twenty one)

[0148] Step 5.6: Use the Adam optimizer to train the time series regression prediction model and calculate the comprehensive loss function To update the model parameters until Until convergence, the optimal retired battery remaining energy estimation model is obtained, which is used to estimate the remaining energy of the retired batteries, so as to estimate the remaining energy of the battery according to the obtained battery remaining energy and the calculated average dynamic internal resistance. Realize battery sorting.

[0149] Step 6: Estimation and sorting of remaining energy of retired power batteries;

[0150] Repeat steps 3-4 for the remaining retired batteries, and only perform low-current hybrid pulse power characteristic (L-HPPC) test and partial charging test during the test phase. After completing feature optimization, deploy the pre-trained residual energy estimation model to estimate the residual energy of each retired battery, and sort the batteries based on the obtained battery residual energy and the corresponding battery average dynamic internal resistance.

[0151] In this embodiment, an electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the above method, and the processor is configured to execute the program stored in the memory.

[0152] In this embodiment, a computer-readable storage medium stores a computer program on the computer-readable storage medium, and the computer program executes the steps of the above method when executed by a processor.

Claims

1. A method for sorting retired power batteries taking into account the imbalanced distribution of battery aging states, characterized in that: The following steps are included: Step 1: Perform appearance screening on a batch of retired power batteries to obtain the remaining retired batteries after screening; Step 2: Randomly select the remaining retired batteries and obtain a total of Test battery of Step 3: Collect test data of the test battery; Step 3.1: After the test battery is left to stand for a period of time, the initial terminal voltage of the test battery is measured. ,in, Representative The terminal voltage of a test battery; Step 3.2: Construct a charging and discharging strategy and use it to Test batteries to get the battery test data set ,in, Indicates Dynamic characteristic data of the test battery, Indicates Static characteristic data of the test battery; Step 4: After preprocessing, the standardized feature sequence is obtained ;in, Represents the The standardized feature sequences of the test batteries; Step 5: Input into LASSO model for processing, and get the A preliminary characterization sequence of test batteries; Step 6: After any two features in the preliminary feature sequence of the test battery are randomly combined, grey correlation analysis is performed to select the features with high correlation coefficient to form the first Optimal feature sequence for a test battery ,in, Representative The kth best feature of the test battery, represents the number of optimal features; Step 7: Build a BiLSTM-based time series regression prediction model, including: feature extraction module, feature smoothing module, residual energy estimation module and temperature-based correction module, and use it to obtain the optimal feature sequence of n test batteries. Processing is performed to obtain the estimated remaining energy of the retired battery ,in, represents the estimated remaining energy of the mth test battery; Step 8: Use formula (8) to obtain the corrected residual energy sequence : (8) In formula (8), is the influence coefficient of temperature on capacity; is the reference temperature; To test the ambient temperature; represents the corrected residual energy value of the mth test battery; Step 9: Construct a comprehensive loss function according to formula (9) : (9) In formula (9), are 2 balance coefficients, represents the loss function of the residual energy estimation, Represents the loss function of the feature smoothing process; Step 10: Use the Adam optimizer to train the time series regression prediction model and calculate the comprehensive loss function To update the model parameters until Until convergence, the optimal retired battery remaining energy estimation model is obtained, which is used to estimate the remaining energy of the retired batteries, so as to estimate the remaining energy of the battery according to the obtained battery remaining energy and the calculated average dynamic internal resistance. Realize battery sorting.

2. A method for sorting retired power batteries considering the imbalanced distribution of battery aging states according to claim 1, characterized in that: Step 3.2 includes the following steps: Step 3.2.1: The low current mixed pulse power characteristics test is performed on the test battery to obtain dynamic characteristic data that can characterize the dynamic characteristics of the battery. ;in, represents the pulse current, and , represents the pulse current of the i-th rate, represents the voltage change during the pulse test, and , It represents the voltage change of the mth test battery in the i-th pulse current, represents the recovery voltage, and , represents the recovery voltage of the mth test cell, represents the time to reach a stable recovery voltage, and , It indicates the time when the mth test battery reaches the stable recovery voltage. Indicates the final voltage at which the test is completed, and ; represents the terminal voltage of the mth test battery; Step 3.2.2 Perform partial charging test on the test battery to obtain static charging data that can characterize the battery aging status information ;in, Indicates the initial state of charge after discharge. Represents the set charging termination state of charge; Indicates the voltage data during the charging process, and , Indicates the charging voltage data of the mth test battery; Indicates the charging capacity data during the charging process, and , Represents the charging capacity data of the mth test battery; Indicates the charging energy data during the charging process, and , Represents the charging energy data of the mth test battery; represents the terminal voltage after partial charging is completed, and , Represents the terminal voltage of the mth test battery.

3. A method for sorting retired power batteries taking into account the imbalanced distribution of battery aging states according to claim 2, characterized in that: Step 3.2.1 is to After the test battery is charged to the initial state charge with a constant current, it is left to stand for a period of time; then the set pulse current is After standing Perform pulse discharge test on each test battery and record the Voltage changes of the test cells during the pulse process ,in, represents the pulse current of the ith rate, Indicates the pulse current of the i-th rate Next The voltage change of the test battery, z represents the number of rates; After the pulse process is completed, The test battery is left to stand until the battery voltage stabilizes, and the corresponding recovery voltage is recorded. , restore the rest time and open circuit voltage .

4. A method for sorting retired power batteries taking into account the imbalanced distribution of battery aging states according to claim 3, characterized in that: Step 3.2.2 is to The test cells were discharged to their initial state of charge , and record the initial voltage at this time , according to the set discharge current ratio After the first test battery is discharged, The test cell is charged at a constant current Charge to the set state of charge , and record the first Test battery voltage , Charging capacity and charging energy After charging is completed, Let the test battery stand for a while and record the Final open circuit voltage of a test cell after partial charging .

5. A method for sorting retired power batteries taking into account the imbalanced distribution of battery aging states according to claim 4, characterized in that: The fourth step comprises: Step 4.1: After data cleaning, we get a clean data set ; Step 4.2: Based on , build a basic feature library ,in, Indicates The dynamic characteristics of the test battery, Indicates A static characterization set of test batteries; Step 4.3: The feature sequence composed of all dynamic features and static features of each test battery and the capacity label of the test battery After splicing and normalization preprocessing, the standardized feature sequence is obtained ,in, Represents the capacity label of the mth tested battery.

6. A method for sorting retired power batteries taking into account the imbalanced distribution of battery aging states according to claim 5, characterized in that: Step 4.2 includes the following steps: Step 4.2.1: The low current mixed pulse power characteristic test data in the image is used for feature extraction to obtain a dynamic feature set. ; Step 4.2.2: The voltage, energy, and capacity data obtained from the partial charging test in the test are averaged, the maximum and minimum extreme values ​​are taken, and used to calculate the variance, kurtosis, skewness, quantile, and slope of the curve to obtain a static feature set. ,in, Indicates The voltage-related characteristics of the test cells, Indicates Capacity-related characteristics of the test cells, Indicates Energy-related characteristics of the test cells, Indicates Partially charged IC curve related characteristics of the tested cells.

7. A method for sorting retired power batteries taking into account the imbalanced distribution of battery aging states according to claim 6, characterized in that: Step 4.2.1 includes the following steps: Step 4.2.1.1: According to Voltage change of the test cell With pulse current Dynamic internal resistance obtained by doing business , using formula (1) and formula (2) to obtain The average dynamic internal resistance of the tested batteries and dynamic internal resistance change rate and accordingly as the First dynamic characterization of the test battery and the second dynamic feature : (1) (2) In formula (1) and formula (2), represents the dynamic internal resistance calculated under the i-th pulse current, and , represents the dynamic internal resistance of the mth test battery under the current pulse current, Indicates the maximum internal resistance, Indicates the minimum value of internal resistance; Step 4.2.1.2: Based on the recovery voltage , recovery time and the initial terminal voltage , using formula (3) and formula (4) to obtain The recovery voltage difference of the tested batteries and pulse recovery rate and accordingly as the The third dynamic characteristics of the test battery and the fourth dynamic feature , thus obtaining the dynamic feature set of n test batteries ; (3) (4) In formula (3) and formula (4), represents the recovery voltage of the mth test battery; represents the recovery time of the mth test battery; Represents the initial terminal voltage of the mth test battery.

8. A method for sorting retired power batteries taking into account the imbalanced distribution of battery aging states according to claim 7, characterized in that: Step 7 includes the following steps: Step 7.1: The feature extraction module uses a BiLSTM layer to Process and output Deep features of the test battery ,in, Indicates Test battery No. deep features, Represents the total number of deep features; Step 7.2: The feature smoothing module is based on the feature distribution smoothing strategy Processing is performed to obtain a smoothed feature sequence ; Step 7.3: The residual energy estimation module is composed of a fully connected layer and a regression estimation layer; wherein a fully connected layer is Perform feature mapping to generate the The regression estimation layer is based on the regression function of the mapping features of the test battery The mapping features of the test battery are processed to generate the Estimated remaining energy of the test battery .

9. A method for sorting retired power batteries taking into account the imbalanced distribution of battery aging states according to claim 8, characterized in that: Step 7.2 includes the following steps: Step 7.2.1: Use equations (5) and (6) to get the updated feature mean and the updated feature variance : (5) (6) In formula (5) and formula (6), represents the smoothing coefficient; and Respectively represent The feature means and feature variances of the test batteries before updating; Step 7.2.2: Use equation (7) to Smoothing is performed to obtain the smoothed feature sequence ; (7) In formula (7), To prevent a tiny constant with a zero denominator; Indicates The first test battery after smoothing A deep feature.

10. A method for sorting retired power batteries taking into account the imbalanced distribution of battery aging states according to claim 9, characterized in that: In step 9, the loss function of the residual energy estimation is obtained by using equations (10) and (11) respectively: And the loss function of the feature smoothing process ; (10) (11)。

Images