Lead-acid storage battery consistency screening method and system based on dynamic and static feature fusion

Abstract

The invention relates to the technical field of decommissioned lead-acid storage battery application, and provides a lead-acid storage battery consistency screening method and system based on dynamic and static feature fusion, and the method comprises the steps: obtaining a plurality of static feature parameters of a decommissioned battery through an EIS test; obtaining a plurality of dynamic characteristic parameters of the retired battery through a pulse test; through feature parameter screening, constructing a feature parameter set fused with static and dynamic feature parameters; adopting a minimum covariance determinant MCD algorithm to improve the Mahalanobis distance, and calculating health scores of all the batteries; adopting an isolation forest unsupervised anomaly detection algorithm to calculate anomaly scores of all batteries; and performing weighted fusion on the health scores and the abnormal scores of the batteries to obtain comprehensive scores of the batteries, and finally screening out a plurality of batteries with highly consistent performance. Through the method, the accuracy and robustness of consistency screening of the retired storage batteries are improved.

Description

Technical Field

[0001] This invention relates to the field of application technology of retired lead-acid batteries, specifically to a consistency screening method and system for lead-acid batteries based on the fusion of dynamic and static features. Background Technology

[0002] The DC power supply system of a substation consists of charging equipment and battery banks. The battery banks provide temporary emergency power during mains power outages, ensuring the continuous normal operation of the charging equipment. Battery banks typically consist of several individual cells connected in series. Inconsistencies in the performance parameters of the individual cells within a bank are a key issue affecting the overall system performance and lifespan. Poor-performing batteries have higher internal resistance, lower capacity, and higher self-discharge rates. If there are batteries with inconsistent performance within the bank, the weakest cell will reach its discharge termination voltage first during a critical moment of mains power interruption. This significantly reduces the effective capacity and discharge time of the entire battery bank, potentially preventing it from reaching its designed backup time and rendering the backup power supply unreliable. Therefore, rigorous screening of individual cell performance consistency is a crucial technical means to improve the reliability and operational safety of uninterruptible power supply systems. It is also a key measure to maximize the actual lifespan of the battery bank, especially in applications primarily using float charging.

[0003] Voltage, capacity, and internal resistance are important indicators for evaluating the consistency of lead-acid batteries. Based on the consistency indicators used, screening methods can be broadly divided into two categories: screening methods based on external characteristic parameters and screening methods based on internal state parameters. Screening methods based on external characteristic parameters directly measure the battery's terminal voltage and temperature, judging the battery's performance consistency based on the differences in terminal voltage and temperature of individual cells within the battery pack under charging, discharging, or resting conditions. While external characteristics are easy to measure, they cannot directly reflect the battery's internal state. Screening methods based on internal state parameters focus on reflecting the battery's internal electrochemical state, such as state of charge, capacity, and internal resistance. These parameters typically cannot be directly measured and require model estimation or specific testing. Electrochemical impedance spectroscopy (EIS) measures the battery's complex impedance response by applying a small-amplitude sinusoidal AC signal over a wide frequency range, providing rich information about internal electrochemical processes such as charge transfer, mass diffusion, and interfacial capacitance. For example, the Chinese invention patent CN120490818A, "A Method for Screening Capacity Consistency of Decommissioned Lead-Acid Batteries Based on Electrochemical Impedance Spectrum Characteristic Parameters," uses measured impedance spectrum curves. After validating the results through KK transformation, some frequency range data are discarded. An equivalent circuit model is established using LR(CR)(QR), and the impedance spectrum is fitted to determine the capacity consistency screening characteristic parameters. However, this method does not consider the dynamic response characteristics of the battery contained in the charge-discharge curves. Screening methods based on dynamic characteristics distinguish the battery state based on the characteristics contained in the charge-discharge curves and their derived curves. Therefore, its screening accuracy is relatively low.

[0004] External characteristic parameters are easy to obtain, but cannot directly reflect the internal state of the battery. Internal characteristics cannot be directly measured and need to be obtained through model estimation or specific tests. However, the discharge curves measured by experiments still contain noise and cannot accurately describe the dynamic characteristics of the battery. The performance of deep learning models is based on a large amount of training data and is not suitable for the problem of real-time consistency screening of battery cells in substations. Summary of the Invention

[0005] The technical problem to be solved by this invention is how to integrate the static and dynamic characteristic parameters of retired lead-acid batteries to improve the accuracy and robustness of the consistency screening of retired batteries.

[0006] The present invention solves the above-mentioned technical problems through the following technical means: This invention provides a consistency screening method for lead-acid batteries based on dynamic and static feature fusion, comprising the following steps: S1. Obtain several static characteristic parameters of retired batteries through EIS testing; S2. Obtain several dynamic characteristic parameters of retired batteries through pulse testing; S3. Calculate the statistical dispersion and correlation of the feature parameters obtained from steps S1 and S2, remove redundant feature parameters, retain representative feature parameters, and construct a feature parameter set that fuses static and dynamic feature parameters. ; S4. Based on the feature parameter set, the minimum covariance determinant (MCD) algorithm is used to improve the Mahalanobis distance, and the health score of all batteries is calculated. ; S5. Based on the feature parameter set, the isolated forest unsupervised anomaly detection algorithm is used to calculate the anomaly score of all batteries. ; S6, Weighted Fusion Battery Health Rating and abnormal scoring To obtain a comprehensive score for the battery. Based on the scoring results, several batteries with highly consistent performance are selected from the retired batteries for recombining and reuse.

[0007] Further, step S1 includes the following steps: S11. Apply a small sinusoidal current perturbation signal to the two ends of the battery, measure the voltage signal of the battery system response, and acquire the current excitation signal. i [ g and voltage response signal u [ g The impedance of the battery at different frequencies is obtained by using mathematical operations such as the Fast Fourier Transform, and the EIS curve is constructed. The impedance is calculated as follows:

[0008] in, Impedance at different frequencies and These are the lead-acid battery terminal voltages. u [ g Fourier transform and lead-acid battery terminal current i [ g The Fourier transform of ] is as follows:

[0009] in, g To change the number of points, j Represents the imaginary number in the complex plane. ω Indicates angular frequency; S12. Construct an equivalent circuit model and extract a set of parameters characterizing the static features of the battery from the EIS curve. L 1, R 0, R 1, Q 1, R 2, Q 2, Z w},in, Q It is a constant-phase element, used in nonlinear least squares fitting. f s , d It consists of two parameters, f s express QR The peak frequency of the circuit d Indicates the constant phase index; Z w yes Warburg Impedance, in nonlinear least squares fitting, is determined by R w , τ , n w It consists of three parameters. R w This represents the characteristic polarization resistance associated with the diffusion process. τ Represents the time constant. n w This represents the diffusion index, with values ​​ranging from [0, 0.5]. R w The calculation formula is as follows:

[0010] in, jIt is the imaginary unit. ω Indicates angular frequency; S13. Obtain the static characteristic parameters of the battery. R 1, R 2, R w .

[0011] Further, step S2 includes the following steps: S21. Apply a brief constant current pulse to the battery and extract dynamic characteristic parameters characterizing the battery under charge and discharge conditions from the short-time pulse charge and discharge experiment, including: discharge dynamic internal resistance. R _ d Charging dynamic internal resistance R _ c Discharge voltage drop Δ V _ d Charging voltage rise Δ V _ c Percentage of voltage recovery 5 seconds after the discharge pulse ends V (5 s )%, pulse voltage curve DTW distance V _ DTW; The calculations of the above dynamic characteristic parameters are as follows:

[0012]

[0013]

[0014]

[0015]

[0016]

[0017] in, v t1 ~ v t7 These represent the voltage values ​​at key time points on the pulse charge-discharge voltage curve. I discharge Indicates the discharge current. I charge Indicates the charging current. v dr_5s This indicates the voltage across the battery 5 seconds after the discharge pulse ends. x 1, x 2, …, x n ] represents the voltage sequence T s1 Vector form, [y 1, y 2, …, y n ] represents the voltage sequence T s2 Vector form; S22. Obtain the dynamic characteristic parameters of the battery. R _ d , R _ c , Δ V _ d ,Δ V_c , V(5s)% , V_DTW .

[0018] Further, step S3 includes the following steps: S31. Quantitatively evaluate the dispersion of each feature parameter using the median, interquartile range (IQR), and coefficient of variation (CV). S32. Verify the redundancy of the characteristic parameters through Spearman correlation analysis; S33. Retain feature parameters with relatively large IQR and CV values, and which exhibit low correlation in both static and dynamic parameters. R 1 and V _ DTW Construct a set of feature parameters , i Indicates the first i Saves battery.

[0019] Further, step S4 includes the following steps: S41. Calculate the set of characteristic parameters Robust mean and covariance, thereby improving Mahalanobis distance; S42. The similarity and aggregation degree between individual cells are evaluated using the improved Mahalanobis distance to obtain the health score of all cells. .

[0020] Further, step S41 includes the following steps: S411, from N Randomly selected from the samples h , denoted by subset index. S k {1,2,…, h The mean and covariance of this subset are calculated as follows:

[0021] Find the subset that has the smallest positive definite determinant. S initThe subset must satisfy the following formula:

[0022] The corresponding initial mean and covariance are calculated as follows:

[0023] in, i * The index corresponding to the optimal subset; S412, Iteratively find the Mahalanobis distance minimum. h 1 sample, until convergence; S413. Output the robust mean and covariance, as follows:

[0024] S414. The improved Mahalanobis distance is calculated as follows: .

[0025] Further, the battery health score mentioned in step S42 The calculation is as follows:

[0026] Where, max( D M * ), min( D M * ) are the maximum and minimum values ​​of the improved Mahalanobis distance for all battery samples, respectively.

[0027] Further, step S5 includes the following steps: S51. Set the maximum height of a single isolation tree to... l =[log2 ψ ],in ψ This represents the number of samples taken from a single isolated tree. S52, From the dataset X Randomly selected from ψ Each sample, from the full set of features F ={ q 1, q 2,…, q p Randomly select 1 feature from} q * In features q * Uniformly and randomly select dividing points within the range of values p * According to characteristicsq * and dividing point p * Divide the samples into the left subset. X l ={ X i ∈ X | q ( X i )< p } and right subset X r ={ X i ∈ X | q ( X i ) ≥ p }; S53. Recursively divide the left and right subsets until the height limit is reached. l If a subset is indivisible, a single isolated tree is obtained. ITree ; S54, Repeat steps S52 and S53. t Next, build an isolated forest. IForest =[ ITree 1, ITree 2, …, ITree t ]; S55. Define the path length function PathLength( X i , ITree , e ),in e The initial path length and at the root node e =0, calculate the first... i The average path length E of the battery in the isolated forest h ( X i ], as shown in the following formula:

[0028]

[0029] S56, Calculate the first i Abnormal scores of battery s ( X i ), as shown in the following formula:

[0030]

[0031] in, c ( n ) indicates in n The average path length of the isolation tree constructed from a battery sample. E ( h ( X i )) indicates the sample battery X i exist t Average path length in the isolated trees; S57. Calculate the anomaly score for all batteries. As shown in the following formula:

[0032] Where, max( s ), min( s ) are the maximum and minimum abnormal scores for all battery samples, respectively.

[0033] Furthermore, the overall score of the battery As shown in the following formula:

[0034] in, W M This represents the weight of the consistency score based on Mahalanobis distance. W IF This represents the weight of the consistency score based on the isolated forest.

[0035] This invention also provides a lead-acid battery consistency screening system based on dynamic and static feature fusion. The system executes the above-described method during operation and includes the following modules: The static feature acquisition module is used to obtain several static feature parameters of retired batteries through EIS testing. The dynamic feature acquisition module is used to obtain several dynamic feature parameters of retired batteries through pulse testing; The feature selection module is used to calculate the statistical dispersion and correlation of feature parameters obtained from the static and dynamic feature acquisition modules, remove redundant feature parameters, retain representative feature parameters, and construct a feature parameter set that fuses static and dynamic feature parameters. ; The health rating module is used to calculate the health rating of all batteries based on the feature parameter set and employs the minimum covariance determinant (MCD) algorithm to improve the Mahalanobis distance. ; The anomaly scoring module is used to calculate the anomaly score of all batteries based on the feature parameter set and employing the isolated forest unsupervised anomaly detection algorithm. ; Output module for weighted fusion battery health scoring and abnormal scoring To obtain a comprehensive score for the battery. Based on the scoring results, several batteries with highly consistent performance are selected from the retired batteries for recombination and reuse. The advantages of this invention are: (1) The multi-stage consistency screening method for lead-acid batteries based on dynamic and static feature fusion proposed in this invention obtains a comprehensive set of dynamic and static feature parameters that characterize the internal state of the battery. Based on the evaluation results of the improved Mahalanobis distance and isolated forest anomaly detection algorithm, the health scores of all batteries are evaluated, thereby achieving rapid screening of lead-acid battery performance consistency. Dynamic and static features characterize the health state of the battery from different dimensions, complementing each other and effectively overcoming the feature bias and evaluation bias problems caused by traditional methods that rely on only a single data source.

[0036] (2) The present invention improves upon the traditional Mahalanobis distance calculation method by using the minimum covariance determinant (MCD) algorithm to calculate the robust mean of the feature data. μ * Covariance Matrix Σ * It reflects the true characteristic distribution of most healthy batteries. Mahalanobis distance is good at identifying pattern bias, that is, the combination of battery characteristics does not conform to the statistical correlation of the healthy group. Even if the individual parameters are not extreme, it focuses on the consistency within the group and improves the robustness of the screening method.

[0037] (3) The isolated forest algorithm used in this invention is good at identifying sparse anomalies, that is, points that exist in isolation in the feature space and are far away from any major data cluster. It focuses on significant deviations from the healthy battery set and combines the two to make a comprehensive health score for the battery, which can capture different types of abnormal batteries. Attached Figure Description

[0038] Figure 1 This is a flowchart illustrating the multi-stage consistency screening method for lead-acid batteries based on dynamic and static feature fusion, as described in an embodiment of the present invention. Figure 2 The Nyquist plot of the original impedance spectrum of a lead-acid battery measured in an embodiment of the present invention; Figure 3 This is a schematic diagram of the equivalent circuit model used to extract internal resistance information from the Nyquist plot of the impedance spectrum according to an embodiment of the present invention; Figure 4This is a schematic diagram of the short-time pulse charge-discharge voltage curve and key time points in an embodiment of the present invention; Figure 5 This is a schematic diagram of the box-line distribution of the feature parameters in an embodiment of the present invention; Figure 6 Spearman correlation heatmap of the nine feature parameters extracted in this embodiment of the invention; Figure 7 This is a visualization of the improved Mahalanobis distance of 104 lead-acid batteries in two-dimensional feature space and a schematic diagram of the aggregation degree of individual cells based on the improved Mahalanobis distance. Figure 8 This is a visualization diagram of the lead-acid battery anomaly score calculated by the isolated forest algorithm in the feature space after PCA dimensionality reduction, according to an embodiment of the present invention. Figure 9 This is a schematic diagram showing the overall score distribution of all batteries in an embodiment of the present invention; Figure 10 A schematic diagram showing the charge / discharge capacity test, open-circuit voltage test, and series discharge test performed on 24 batteries selected in this embodiment of the invention. Figure 11 This is a schematic diagram showing the results of a series discharge test on 24 batteries selected in this embodiment of the invention. Figure 12 This is a schematic diagram comparing the capacity of the 24 batteries selected in the embodiments and comparative examples of the present invention in terms of the capacity of a single battery cell. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Example 1 In this embodiment, the test subjects are 104 retired lead-acid batteries. Each battery has a rated voltage of 2V, a nominal capacity of 200Ah, a floating charge voltage range of 2.23V to 2.27V at 25°C, an equalization charge voltage range of 2.30V to 2.35V, and a maximum allowable charging current of 60A. Using the lead-acid battery consistency screening method based on dynamic and static feature fusion provided in this embodiment, 24 lead-acid batteries with good consistency are finally connected in series for further application in scenarios with less stringent battery performance requirements. The specific implementation process is as follows... Figure 1 As shown, it includes the following steps: S1. Obtain several static characteristic parameters of the retired battery through EIS testing; the specific implementation method includes the following steps: S11. Apply a small sinusoidal current perturbation signal to the two ends of the battery, measure the voltage signal of the battery system response, and acquire the current excitation signal. i [ g and voltage response signal u [ g Using mathematical operations such as Fast Fourier Transform, the impedance of the battery at different frequencies is obtained, and EIS curves are constructed, such as... Figure 2 As shown, the EIS curve of a lead-acid battery consists of three parts, which correspond to different electrochemical processes inside the battery: the inductive effect in the high-frequency region, the charge transfer process and the double-layer capacitance effect in the mid-frequency region, and the diffusion process in the low-frequency region. The impedance is calculated as follows:

[0041] in, Impedance at different frequencies and These are the lead-acid battery terminal voltages. u [ g Fourier transform and lead-acid battery terminal current i [ g The Fourier transform of ] is as follows:

[0042] in, g To change the number of points, j Represents the imaginary number in the complex plane. ω Indicates angular frequency; S12. Construct an equivalent circuit model, such as Figure 3 As shown, a set of parameters characterizing the static features of the battery is extracted from the EIS curve. L 1, R 0, R 1, Q 1, R 2, Q 2, Z w},in, Q It is a constant-phase element, used in nonlinear least squares fitting. f s , d It consists of two parameters, f s express QR The peak frequency of the circuit d Indicates the constant phase index; Z w yes WarburgImpedance, in nonlinear least squares fitting, is determined by R w , τ , n w It consists of three parameters. R w This represents the characteristic polarization resistance associated with the diffusion process. τ Represents the time constant. n w This represents the diffusion index, with values ​​ranging from [0, 0.5]. R w The calculation formula is as follows:

[0043] in, j It is the imaginary unit. ω Indicates angular frequency; S13. Obtain the static characteristic parameters of the battery. R 1, R 2, R w .

[0044] S2. Obtain several dynamic characteristic parameters of the retired battery through pulse testing; the specific implementation method includes the following steps: S21. Apply a brief constant current pulse to the battery and extract dynamic characteristic parameters characterizing the battery under charge and discharge conditions from the short-time pulse charge and discharge experiment, including: discharge dynamic internal resistance. R _ d Charging dynamic internal resistance R _ c Discharge voltage drop Δ V _ d Charging voltage rise Δ V _ c Percentage of voltage recovery 5 seconds after the discharge pulse ends V (5 s )%, pulse voltage curve DTW distance V _ DTW; like Figure 4 The key time points on the short-time pulse charge-discharge voltage curve shown are used to calculate the set of dynamic characteristic parameters.

[0045] The pulse test procedure is as follows: (1) Initialize the battery by letting it rest for 10 seconds and record the voltage during the resting period; (2) Apply a constant current negative pulse to the battery. The current is 0.2C (40A) and the pulse duration is 10s. If the battery voltage drops to the discharge cutoff voltage of 1.8V within 10s, the pulse will stop automatically. Record the voltage response during this pulse. (3) After the negative pulse is completed or stopped, let the battery rest for 10 seconds and record the voltage at both ends of the battery during the resting period; (4) Apply a positive pulse with a constant current of 0.2C (40A) for 10s. If the voltage across the battery reaches the 2.35V charging cutoff voltage within 10s, the pulse will stop automatically. Record the voltage response during this pulse. (5) The test is completed after the positive pulse ends.

[0046] The calculations of the above dynamic characteristic parameters are as follows:

[0047]

[0048]

[0049]

[0050]

[0051]

[0052] in, v t1 ~ v t7 These represent the voltage values ​​at key time points on the pulse charge-discharge voltage curve. I discharge Indicates the discharge current. I charge Indicates the charging current. v dr_5s This indicates the voltage across the battery 5 seconds after the discharge pulse ends. x 1, x 2, …, x n ] represents the voltage sequence T s1 Vector form, [ y 1, y 2, …, y n ] represents the voltage sequence T s2 Vector form; S22. Obtain the dynamic characteristic parameters of the battery. R _ d , R _ c, Δ V _ d ,Δ V_c , V(5s)% , V_DTW .

[0053] Based on the nine feature parameters obtained from the above steps, such as Figure 5 The box plots showing the distribution of the characteristic parameters illustrate the distribution of nine characteristic parameters across all batteries through nine subplots. The plots reveal that the ohmic internal resistance and low-frequency impedance real part characteristic parameters extracted from EIS experimental data exhibit a flattened box shape with significant outliers. This indicates that most batteries show good consistency in these characteristic parameters, and any obvious outlier is highly likely to indicate a problematic battery. In contrast, the transfer impedance characteristic parameter extracted from EIS experimental data, the dynamic charging resistance extracted from pulse experimental data, and the instantaneous voltage drop during discharge pulses exhibit very high box shapes, suggesting significant individual differences in these kinetic-related characteristics among batteries.

[0054] S3. Calculate the statistical dispersion and correlation of the feature parameters obtained from steps S1 and S2, remove redundant feature parameters, retain representative feature parameters, and construct a feature parameter set that fuses static and dynamic feature parameters. The specific implementation method includes the following steps: S31. The dispersion of each feature parameter is quantitatively evaluated using the median, interquartile range (IQR), and coefficient of variation (CV); among which, R 1 and V _ DTW The relatively large IQR and CV values ​​indicate significant differences between battery cells, thus possessing strong differentiation potential. In contrast, R 0、 R w and R _ d The parameters included have a very narrow IQR range, indicating that their ability to distinguish battery states is limited.

[0055] S32. Verify the redundancy of characteristic parameters using Spearman correlation analysis; such as Figure 6 As shown, there is a strong correlation among the resistance characteristics derived from multiple pulses, for example... R _ d and Δ V _ d ( ρ = 0.999), and R _ c and Δ V _ c ( ρ=0.644), indicating a significant overlap in the information they convey. Furthermore, V (5 s )%and V _ DTW There is a perfect correlation ( ρ = 1.000), indicating that these two characteristics are interchangeable. In contrast, the charge transfer impedance derived from electrochemical impedance spectroscopy... R 1 has consistently shown low correlation with other features (| ρ |<0.15), indicating that this feature is relatively independent. Similarly, V _ DTW The correlation with most impedance and resistance-based parameters is weak, reflecting complementary information related to the dynamic voltage evolution during pulsed discharge. Therefore, considering the high dispersion and low redundancy, R 1 and V _ DTW Ultimately, it was selected as the core feature for battery consistency screening, while the other parameters were excluded due to excessive redundancy or insufficient discrimination ability.

[0056] S33. Retain feature parameters with relatively large IQR and CV values, and which exhibit low correlation in both static and dynamic parameters. R 1 and V _ DTW Construct a set of feature parameters , i Indicates the first i Saves battery.

[0057] After determining the set of characteristic parameters for dynamic and static fusion, the comprehensive score of all batteries is calculated.

[0058] S4. The mean μ and covariance matrix Σ obtained by the traditional Mahalanobis distance algorithm are highly sensitive to outliers. During battery screening, a few batteries with early failures or manufacturing defects may significantly affect the estimation of the center and dispersion of healthy battery packs, leading to inaccurate Mahalanobis distance calculations and masking truly abnormal batteries. Therefore, based on the feature parameter set, the minimum covariance determinant (MCD) algorithm is used to improve the Mahalanobis distance calculation and calculate the health score of all batteries. Its core idea is: to find a subset of data in the original dataset that contains... h The method for finding a subset of data points that minimizes the determinant of the covariance matrix calculated from that subset includes the following steps: S41. Calculate the set of characteristic parameters The robust mean and covariance are used to improve the Mahalanobis distance; the specific improvement method includes the following steps: S411, from N Randomly selected from the samples h , denoted by subset index. S k {1,2,…, h The mean and covariance of this subset are calculated as follows:

[0059] Find the subset that has the smallest positive definite determinant. S init The subset must satisfy the following formula:

[0060] The corresponding initial mean and covariance are calculated as follows:

[0061] in, i * The index corresponding to the optimal subset; S412, Iteratively find the Mahalanobis distance minimum. h 1 sample, until convergence; S413. Output the robust mean and covariance, as follows:

[0062] S414. The improved Mahalanobis distance is calculated as follows: .

[0063] S42. Utilizing the improved Mahalanobis distance to evaluate the similarity and aggregation degree among battery cells, the improved Mahalanobis distance... D M * Standardize the data, subtract the standardized distance from 1, and obtain the health score for all batteries. The specific calculation is as follows:

[0064] Where, max( D M * ), min( D M * ) are the maximum and minimum values ​​of the improved Mahalanobis distance for all battery samples, respectively.

[0065] like Figure 7As shown, most cells cluster in the central region of the feature space, indicating that they have similar electrochemical properties and good consistency; however, a few cells deviate significantly from the central cluster, with their Mahalanobis distance being significantly higher than that of the other cells. For example, Figure 7 In (a), the lighter the color of batteries such as Battery_16 and Battery_48, the larger the Mahalanobis distance, indicating that they are more likely to be statistical outliers in the battery pack. This may be due to reasons such as excessively high dynamic charging resistance or excessively high charge transfer impedance. Battery_98 has a Mahalanobis distance of more than 101, and is also an outlier. Figure 7 (b) The darker the color, the higher the battery score, the closer the battery is to the group center, and the better the consistency performance.

[0066] S5. Based on the feature parameter set, the isolated forest unsupervised anomaly detection algorithm is used to calculate the anomaly score of all batteries. During the training phase, a subset of samples and features are extracted from the original dataset to construct isolation trees. Outliers are a minority of points that are far from normal points in the feature space, making them easier to isolate through random partitioning. Utilizing the idea of ​​ensemble learning, samples and features are extracted multiple times to construct multiple isolation trees until the data is no longer divisible. This stage includes two parameters: t Indicates the number of trees. ψ This indicates that each tree is constructed from the set of feature parameters. A randomly selected subset of data. The specific implementation includes the following steps: S51. Set the maximum height of a single isolation tree to... l =[log2 ψ ],in ψ This represents the number of samples taken from a single isolated tree. S52, From the dataset X Randomly selected from ψ Each sample, from the full set of features F ={ q 1, q 2,…, q p Randomly select 1 feature from} q * In features q * Uniformly and randomly select dividing points within the range of values p * According to characteristics q * and dividing point p * Divide the samples into the left subset. Xl ={ X i ∈ X | q ( X i )< p } and right subset X r ={ X i ∈ X | q ( X i ) ≥ p }; S53. Recursively divide the left and right subsets until the height limit is reached. l If a subset is indivisible, a single isolated tree is obtained. ITree ; S54, Repeat steps S52 and S53. t Next, build an isolated forest. IForest =[ ITree 1, ITree 2, …, ITree t ]; S55. Define the path length function PathLength( X i , ITree , e ),in e The initial path length and at the root node e =0, calculate the first... i The average path length E of the battery in the isolated forest h ( X i ], as shown in the following formula:

[0067]

[0068] S56, Calculate the first i Abnormal scores of battery s ( X i ), as shown in the following formula:

[0069]

[0070] in, c ( n ) indicates inn The average path length of the isolation tree constructed from a battery sample. E ( h ( X i )) indicates the sample battery X i exist t Average path length in the isolated trees; S57. Calculate the anomaly score for all batteries. As shown in the following formula:

[0071] Where, max( s ), min( s ) are the maximum and minimum abnormal scores for all battery samples, respectively.

[0072] like Figure 8 As shown, darker colors indicate a lower degree of anomalousness for the battery, while lighter colors indicate a higher degree of anomalousness. The results show that most batteries have low anomalousness scores, indicating that the model considers them normal samples. A few batteries, however, received high anomalousness scores, and the model judged them as anomalous. Batteries marked as highly anomalous by I (e.g., Battery_16, Battery_58) overlap with batteries from the previous section that are far from Mahalanobis. For example, Battery_16 and Battery_58 exhibited extremely high anomalousness in both methods.

[0073] S6, Weighted Fusion Battery Health Rating and abnormal scoring To obtain a comprehensive score for the battery. Based on the scoring results, several batteries with highly consistent performance are selected from the retired batteries for recombining and reuse.

[0074] The battery's overall rating As shown in the following formula:

[0075] in, W M This represents the weight of the consistency score based on Mahalanobis distance. W IF This represents the weight of the consistency score based on the isolated forest.

[0076] like Figure 9 As shown, most batteries scored in the high range of 0.8 or above, but there is a clear long tail, with some batteries scoring far below the group average. These batteries are identified as anomalous batteries.

[0077] By calculating the overall health score of all individual cells Score total The 24 batteries with the highest scores were selected as the battery pack with good consistency, and charge / discharge capacity, open-circuit voltage, and series discharge tests were performed on these 24 batteries. Figure 10 As shown in (a), the capacity of the 24 batteries is distributed in the range of 227~232Ah, with a capacity difference of 2.11%; Figure 10 As shown in (b), the open-circuit voltage difference of the 24 batteries before discharge is 16mV. Figure 11 As shown, all 24 batteries, after being assembled, underwent a complete 10-hour discharge process. The voltage difference between the individual batteries reached its maximum at the 10th hour, but was only 73mV. The test results for battery capacity, open-circuit voltage, and series discharge fully demonstrate that the method provided in this embodiment effectively screened 24 batteries from 104 retired batteries, exhibiting excellent consistency and high accuracy.

[0078] Example 2 It should be further explained that, based on the same inventive concept, this embodiment provides a lead-acid battery consistency screening system based on dynamic and static feature fusion. When the system is running, it executes the method described in Embodiment 1, including the following modules: The static feature acquisition module is used to obtain several static feature parameters of retired batteries through EIS testing. The dynamic feature acquisition module is used to obtain several dynamic feature parameters of retired batteries through pulse testing; The feature selection module is used to calculate the statistical dispersion and correlation of feature parameters obtained from the static and dynamic feature acquisition modules, remove redundant feature parameters, retain representative feature parameters, and construct a feature parameter set that fuses static and dynamic feature parameters. ; The health rating module is used to calculate the health rating of all batteries based on the feature parameter set and employs the minimum covariance determinant (MCD) algorithm to improve the Mahalanobis distance. ; The anomaly scoring module is used to calculate the anomaly score of all batteries based on the feature parameter set and employing the isolated forest unsupervised anomaly detection algorithm. ; Output module for weighted fusion battery health scoring and abnormal scoring To obtain a comprehensive score for the battery. Based on the scoring results, several batteries with highly consistent performance are selected from the retired batteries for recombining and reuse.

[0079] Comparative Example In this comparative example, 104 retired lead-acid batteries, identical to those used in the previous example, were employed. Each battery had a rated voltage of 2V, a nominal capacity of 200Ah, a floating charge voltage range of 2.23V~2.27V at 25°C, an equalization charge voltage range of 2.30V~2.35V, and a maximum allowable charging current of 60A. A health assessment score was calculated for each battery directly using the set of characteristic parameters extracted from EIS testing and short-time pulse charge-discharge experiments. The main steps are as follows: Step 1: Perform electrochemical impedance spectroscopy and short-time pulse charge-discharge experiments to obtain a set of characteristic parameters characterizing the internal health state of the battery; Step 2: Select parameters that are highly correlated with the battery health status from the set of feature parameters, and use the entropy weight method to calculate a health assessment score for each battery. Step 3: Select the 24 batteries with the highest health assessment scores for verification.

[0080] like Figure 12 As shown, compared with the weighted evaluation method using feature parameters, the consistency screening method proposed in Example 1 selects 24 batteries with better consistency.

[0081] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A consistency screening method for lead-acid batteries based on dynamic and static feature fusion, characterized in that, Includes the following steps: S1. Obtain several static characteristic parameters of retired batteries through EIS testing; S2. Obtain several dynamic characteristic parameters of retired batteries through pulse testing; S3. Calculate the statistical dispersion and correlation of the feature parameters obtained from steps S1 and S2, remove redundant feature parameters, retain representative feature parameters, and construct a feature parameter set that fuses static and dynamic feature parameters. ; S4. Based on the feature parameter set, the minimum covariance determinant (MCD) algorithm is used to improve the Mahalanobis distance, and the health score of all batteries is calculated. ; S5. Based on the feature parameter set, the isolated forest unsupervised anomaly detection algorithm is used to calculate the anomaly score of all batteries. ; S6, Weighted Fusion Battery Health Rating and abnormal scoring To obtain a comprehensive score for the battery. Based on the scoring results, several batteries with highly consistent performance are selected from the retired batteries for recombining and reuse.

2. The lead-acid battery consistency screening method based on dynamic and static feature fusion according to claim 1, characterized in that, Step S1 includes the following steps: S11. Apply a small sinusoidal current perturbation signal to the two ends of the battery, measure the voltage signal of the battery system response, and acquire the current excitation signal. i [ g and voltage response signal u [ g The impedance of the battery at different frequencies is obtained by using mathematical operations such as the Fast Fourier Transform, and the EIS curve is constructed. The impedance is calculated as follows: in, Impedance at different frequencies and These are the lead-acid battery terminal voltages. u [ g Fourier transform and lead-acid battery terminal current i [ g The Fourier transform of ] is as follows: in, g To change the number of points, j Represents the imaginary number in the complex plane. ω Indicates angular frequency; S12. Construct an equivalent circuit model and extract a set of parameters characterizing the static features of the battery from the EIS curve. L 1, R 0, R 1, Q 1, R 2, Q 2, Z w },in, Q It is a constant-phase element, used in nonlinear least squares fitting. f s , d It consists of two parameters, f s express QR The peak frequency of the circuit d Indicates the constant phase index; Z w yes Warburg Impedance, in nonlinear least squares fitting, is determined by R w , τ , n w It consists of three parameters. R w This represents the characteristic polarization resistance associated with the diffusion process. τ Represents the time constant. n w This represents the diffusion index, with values ​​ranging from [0, 0.5]. R w The calculation formula is as follows: in, j It is the imaginary unit. ω Indicates angular frequency; S13. Obtain the static characteristic parameters of the battery. R 1, R 2, R w .

3. The lead-acid battery consistency screening method based on dynamic and static feature fusion according to claim 2, characterized in that, Step S2 includes the following steps: S21. Apply a brief constant current pulse to the battery and extract dynamic characteristic parameters characterizing the battery under charge and discharge conditions from the short-time pulse charge and discharge experiment, including: discharge dynamic internal resistance. R _ d Charging dynamic internal resistance R _ c Discharge voltage drop Δ V _ d Charging voltage rise Δ V _ c Percentage of voltage recovery 5 seconds after the discharge pulse ends V (5 s )%, pulse voltage curve DTW distance V _ DTW; The calculations of the above dynamic characteristic parameters are as follows: in, v t1 ~ v t7 These represent the voltage values ​​at key time points on the pulse charge-discharge voltage curve. I discharge Indicates the discharge current. I charge Indicates the charging current. v dr_5s This indicates the voltage across the battery 5 seconds after the discharge pulse ends. x 1, x 2, …, x n ] represents the voltage sequence T s1 Vector form, [ y 1, y 2, …, y n ] represents the voltage sequence T s2 Vector form; S22. Obtain the dynamic characteristic parameters of the battery. R _ d , R _ c , Δ V _ d , Δ V_c , V(5s)% , V_DTW .

4. The lead-acid battery consistency screening method based on dynamic and static feature fusion according to claim 3, characterized in that, Step S3 includes the following steps: S31. Quantitatively evaluate the dispersion of each feature parameter using the median, interquartile range (IQR), and coefficient of variation (CV). S32. Verify the redundancy of the characteristic parameters through Spearman correlation analysis; S33. Retain feature parameters with relatively large IQR and CV values, and which exhibit low correlation in both static and dynamic parameters. R 1 and V _ DTW Construct a set of feature parameters , i Indicates the first i Saves battery.

5. The lead-acid battery consistency screening method based on dynamic and static feature fusion according to claim 1, characterized in that, Step S4 includes the following steps: S41. Calculate the set of characteristic parameters Robust mean and covariance, thereby improving Mahalanobis distance; S42. The similarity and aggregation degree between individual cells are evaluated using the improved Mahalanobis distance to obtain the health score of all cells. .

6. The lead-acid battery consistency screening method based on dynamic and static feature fusion according to claim 5, characterized in that, Step S41 includes the following steps: S411, from N Randomly selected from the samples h , denoted by subset index. S k {1,2,…, h The mean and covariance of this subset are calculated as follows: Find the subset that has the smallest positive definite determinant. S init The subset must satisfy the following formula: The corresponding initial mean and covariance are calculated as follows: in, i * The index corresponding to the optimal subset; S412, Iteratively find the Mahalanobis distance minimum. h 1 sample, until convergence; S413. Output the robust mean and covariance, as follows: S414. The improved Mahalanobis distance is calculated as follows: 。 7. The lead-acid battery consistency screening method based on dynamic and static feature fusion according to claim 6, characterized in that, The battery health score described in step S42 The calculation is as follows: Where, max( D M * ), min( D M * ) are the maximum and minimum values ​​of the improved Mahalanobis distance for all battery samples, respectively.

8. The lead-acid battery consistency screening method based on dynamic and static feature fusion according to claim 1, characterized in that, Step S5 includes the following steps: S51. Set the maximum height of a single isolation tree to... l =[log2 ψ ],in ψ This represents the number of samples taken from a single isolated tree. S52, From the dataset X Randomly selected from ψ Each sample, from the full set of features F ={ q 1, q 2,…, q p Randomly select 1 feature from} q * In features q * Uniformly and randomly select dividing points within the range of values p * According to characteristics q * and dividing point p * Divide the samples into the left subset. X l ={ X i ∈ X | q ( X i ) < p } and right subset X r ={ X i ∈ X | q ( X i ) ≥ p }; S53. Recursively divide the left and right subsets until the height limit is reached. l If a subset is indivisible, a single isolated tree is obtained. ITree ; S54, Repeat steps S52 and S53. t Next, build an isolated forest. IForest =[ ITree 1, ITree 2, …, ITree t ]; S55. Define the path length function PathLength( X i , ITree , e ),in e The initial path length and at the root node e =0, calculate the first... i The average path length E of the battery in the isolated forest h ( X i ], as shown in the following formula: S56, Calculate the first i Abnormal scores of battery s ( X i ), as shown in the following formula: in, c ( n ) indicates in n The average path length of the isolation tree constructed from a battery sample. E ( h ( X i )) indicates the sample battery X i exist t Average path length in the isolated trees; S57. Calculate the anomaly score for all batteries. As shown in the following formula: Where, max( s ), min( s ) are the maximum and minimum abnormal scores for all battery samples, respectively.

9. The lead-acid battery consistency screening method based on dynamic and static feature fusion according to claim 1, characterized in that, The battery's overall rating As shown in the following formula: in, W M This represents the weight of the consistency score based on Mahalanobis distance. W IF This represents the weight of the consistency score based on the isolated forest.

10. A lead-acid battery consistency screening system based on dynamic and static feature fusion, characterized in that, Includes the following modules: The static feature acquisition module is used to obtain several static feature parameters of retired batteries through EIS testing. The dynamic feature acquisition module is used to obtain several dynamic feature parameters of retired batteries through pulse testing; The feature selection module is used to calculate the statistical dispersion and correlation of feature parameters obtained from the static and dynamic feature acquisition modules, remove redundant feature parameters, retain representative feature parameters, and construct a feature parameter set that fuses static and dynamic feature parameters. ; The health rating module is used to calculate the health rating of all batteries based on the feature parameter set and employs the minimum covariance determinant (MCD) algorithm to improve the Mahalanobis distance. ; The anomaly scoring module is used to calculate the anomaly score of all batteries based on the feature parameter set and employing the isolated forest unsupervised anomaly detection algorithm. ; Output module for weighted fusion battery health scoring and abnormal scoring To obtain a comprehensive score for the battery. Based on the scoring results, several batteries with highly consistent performance are selected from the retired batteries for recombining and reuse.

Images