Method for estimating health state of retired lead-acid storage battery based on SOM-GA-XGBoost hybrid model

Abstract

The invention discloses a SOM-GA-XGBoost hybrid model-based ex-service lead-acid storage battery state-of-health estimation method, and relates to the field of ex-service lead-acid storage battery state-of-health estimation. By measuring an electrochemical impedance spectrum of an ex-service lead-acid storage battery, an appropriate equivalent circuit model is established according to the curve shape of the electrochemical impedance spectrum; fitting the electrochemical impedance spectrum by using an equivalent circuit model to obtain corresponding equivalent circuit element parameters; dynamic response parameters of the lead-acid storage battery are obtained through a mixed pulse power characteristic test; and establishing an SOM-GA-XGBoost hybrid model, and exploring the corresponding relationship between the characteristic parameters obtained by the experiment and the residual capacity of the decommissioned lead-acid storage battery to realize the health state estimation of the decommissioned lead-acid storage battery. According to the method, the characteristic parameters related to the health state of the decommissioned lead-acid storage battery are obtained from the performance test of the lead-acid storage battery, the residual service life trend of the lead-acid storage battery is more accurately captured, and the predicted value of the model is basically consistent with the change of the true value.

Description

Technical Field

[0001] The present invention relates to the field of health state estimation of retired lead-acid batteries, and in particular to a health state estimation method for retired lead-acid batteries based on a SOM-GA-XGBoost hybrid model. Background Art

[0002] Lead-acid batteries, as a mature energy storage device, are widely used in automotive starting batteries, power system energy storage, solar photovoltaic energy storage systems, and other fields. As lead-acid batteries age, their performance gradually declines. Predicting their remaining useful life is crucial for ensuring the operational reliability and safety of the equipment. The battery's state of health (SOH) is a key indicator of its remaining performance and lifespan, typically expressed as a percentage of its current capacity or performance relative to its rated value when new. For retired lead-acid batteries, accurately assessing their SOH not only determines their suitability for second-life use but also provides guidance for recycling and reuse.

[0003] At present, the main technologies for judging the SOH of retired lead-acid batteries are divided into five categories: technology based on on-site rapid diagnosis, technology based on retirement standards and historical usage data, technology based on electrochemical testing, technology based on electrochemical modeling and parameter estimation, and data-driven intelligent analysis technology. Fast pulse testing uses short-duration pulse charge and discharge tests lasting from a few seconds to tens of seconds to obtain the battery's voltage and current response, allowing for a rapid estimation of the battery's state of health (SOH). This method is fast and suitable for rapid on-site assessments, but the accuracy of the assessment results needs further improvement. Non-contact testing uses non-contact technologies such as infrared thermal imaging and ultrasonic testing to detect the battery's internal condition without disassembling the battery, making it suitable for screening retired batteries. However, this method relies on specific equipment and has significant uncertainty in the diagnostic results. The SOH prediction method, which combines the manufacturer's standard life curve with the battery's cumulative charge and discharge cycles, depth of discharge, and temperature, is simple and easy to use and suitable for large-scale assessments. However, it does not account for variations in the internal state of battery cells during operation, leading to large errors in the estimation results. Verification discharge is a common method for directly measuring a battery's actual available capacity. By fully discharging the battery to a specified cutoff voltage, the total charge it can provide under specific operating conditions is calculated. This method is a direct measurement method that accurately reflects the battery's current available capacity. However, verification discharge requires the battery to be fully discharged to the cutoff voltage, which typically takes a long time, especially for large-capacity batteries. This requirement makes it unsuitable for online use scenarios such as vehicles and power storage systems. To overcome the shortcomings of check discharge, alternative technologies have been proposed. Research has found that internal resistance is closely correlated with battery aging and can be used as a common indicator for determining battery SOH. Common methods include measuring impedance at different frequencies using the AC impedance method, constructing an equivalent circuit model of the battery using equivalent circuit elements, fitting the battery's dynamic behavior, and determining the battery's SOH based on model parameters. The DC internal resistance measurement method uses short-term current pulses to measure voltage changes and derive internal resistance. With the development of machine learning technology, a large number of studies have used historical battery data, such as current, voltage, temperature, and discharge curves, to train regression or classification models to estimate SOH. Common algorithms include support vector machines, decision trees, random forests, and neural networks. These methods are highly adaptable and can capture complex nonlinear relationships, but they rely heavily on data quality and quantity, limiting model generalization. Deep learning algorithms, such as convolutional neural networks, recurrent neural networks, or autoencoders, extract features from raw signals and estimate battery SOH. While these methods can automate feature extraction and process complex, high-dimensional experimental data, they require significant computing resources and data annotation, resulting in poor real-time performance.

[0004] Estimating the remaining state of health (SOH) of retired lead-acid batteries is not only an important means to promote efficient resource utilization and environmental protection, but also a key guarantee for improving economic benefits and battery safety. This technology plays a key role in battery second-life utilization, resource recovery, environmental protection, and industrial chain optimization, and has become an important technical support for achieving sustainable development. However, the factors affecting the health state of retired lead-acid batteries are coupled with each other. How to accurately determine the remaining state of health of retired lead-acid batteries is an urgent problem that needs to be solved. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for estimating the health status of retired lead-acid batteries based on a SOM-GA-XGBoost hybrid model. The method measures the electrochemical impedance spectrum of retired lead-acid batteries and performs a hybrid pulse power characteristic test. The former calculates the impedance through signal acquisition and FFT, and the latter is completed by static and pulse injection. The impedance spectrum is then fitted to obtain an equivalent circuit parameter set, and the test data is analyzed to obtain a performance parameter set. Subsequently, a SOM-GA-XGBoost hybrid model is constructed, and the parameter set is clustered and mapped to a two-dimensional grid using SOM. GA is used to optimize the optimal feature combination. Finally, the optimized features are input into XGBoost training, and the model is evaluated using a test set to estimate the remaining service life of the battery, thereby achieving health status estimation and solving the above-mentioned problems.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for estimating the state of health of retired lead-acid batteries based on a SOM-GA-XGBoost hybrid model, comprising:

[0008] Step 1: Use an electrochemical workstation to measure the electrochemical impedance spectroscopy of retired lead-acid batteries;

[0009] Step 2: Perform a hybrid pulse power characteristic test on the retired lead-acid battery in a charge and discharge tester;

[0010] Step 3: Fit the electrochemical impedance spectrum of the retired lead-acid battery to obtain an equivalent circuit parameter set that characterizes the internal performance characteristics of the retired lead-acid battery; analyze the hybrid pulse power characteristic test data to obtain a performance parameter set that characterizes the internal state of the retired lead-acid battery in the power grid;

[0011] Step 4: Establish a SOM-GA-XGBoost hybrid model to determine the characteristic parameters related to the remaining health state of the battery and realize the health state estimation of retired lead-acid batteries.

[0012] To optimize the above technical solutions, specific measures taken also include:

[0013] Furthermore, the test objects of the present invention are 104 grid-retired "Shuangdeng" GFM series valve-regulated sealed lead-acid batteries, which have a nominal voltage of 2V, a nominal capacity of 500Ah, a charging current of 50A, a float charge voltage of 2.23V, and an equalization charge voltage of 2.3V.

[0014] Furthermore, in step 1, the electrochemical impedance spectroscopy experiment uses a Coster CS2350H electrochemical workstation, the impedance test frequency range is selected to be 0.01 Hz-1 kHz, and the calculation process of the measured impedance is:

[0015] The current excitation signal i[n] and the voltage response signal u[n] are collected and FFT (Fast Fourier Transform) is performed to obtain the frequency domain signals I(jω) and U(jω):

[0016]

[0017] Where i[n] is the current excitation signal, u[n] is the voltage response signal, ω represents the angular frequency, n represents the number of discrete signal points, and j is the imaginary unit;

[0018] Calculate the impedance Z(jω) of the battery under test based on the frequency domain signals I(jω) and U(jω):

[0019] Z(jω)=U(jω) / I(jω)=Z′+Z″

[0020] Wherein, I(jω) and U(jω) are the frequency domain signals calculated by the above formula, and Z′ and Z″ represent the real impedance and imaginary impedance of the battery to be tested, respectively.

[0021] Furthermore, in step 2, a mixed pulse power characteristic test of 104 retired lead-acid batteries from the power grid is performed in a charge and discharge tester. The experimental steps are as follows: the retired lead-acid batteries are allowed to rest for thirty minutes; after resting, the batteries are charged and discharged in the charge and discharge tester at a 0.1C pulse rate, with one pulse lasting 10 seconds; and the batteries are charged and discharged in the charge and discharge tester at a 0.2C pulse rate, with one pulse lasting 10 seconds.

[0022] Furthermore, in step 3, a suitable equivalent circuit model is established based on the measured electrochemical impedance spectroscopy curve. The equivalent circuit model is: L1R0(Q1R1)(Q2R2)Rw. The equivalent circuit model is used to fit the electrochemical impedance spectra of 104 retired lead-acid batteries from the power grid to obtain the equivalent circuit element parameter set: EIS_fs={R0, L1, R1, fs1, n1, R2, fs2, n2, Rw, tau_w, n_w}.

[0023] Furthermore, in step 3, the hybrid pulse power characteristic test data is analyzed to obtain a performance parameter set characterizing the internal state of the retired lead-acid battery in the power grid: HPPC_fs={V drop ,R in ,P}, where V drop It represents the voltage drop after a high-frequency pulse response. Its calculation formula is as follows:

[0024] V drop =V max -V min

[0025] Where V max Indicates the maximum voltage across the battery in a high-frequency pulse response, V min Indicates the minimum voltage across the battery within a high-frequency pulse response.

[0026] Furthermore, the parameter set R in Represents the internal resistance of the battery, which is calculated as follows:

[0027]

[0028] Where ΔV represents the change in voltage across the battery during a high-frequency pulse response, and ΔI represents the change in current inside the battery during a high-frequency pulse response.

[0029] Furthermore, P in the parameter set represents the power response. During a high-frequency pulse response, the output power of the battery is calculated by the following formula:

[0030] P(t)=V(t)×I(t)

[0031] Where V(t) represents the voltage response during a high-frequency pulse, and I(t) represents the current response during a high-frequency pulse.

[0032] Furthermore, in step 4, a SOM-GA-XGBoost hybrid model is established to determine characteristic parameters related to the remaining state of health of the battery, thereby realizing the state of health estimation of retired lead-acid batteries, including the following sub-steps:

[0033] Step 401: using a self-organizing map unsupervised learning method to perform similarity clustering on a parameter set characterizing battery performance, find data points with similar characteristics, and map them onto a two-dimensional grid;

[0034] Step 402: After the SOM performs similarity clustering on the parameters, a genetic algorithm is used to perform heuristic optimization to explore different combinations of performance parameters and find the best combination that can predict the remaining battery life;

[0035] Step 403: Input the features optimized by the genetic algorithm into XGBoost for training, and use the powerful regression ability of XGBoost to estimate the remaining service life of retired lead-acid batteries.

[0036] Furthermore, in step 401, SOM is an unsupervised learning method used to map high-dimensional data into a low-dimensional space while maintaining the topological structure of the data. Unlike traditional neural networks, SOM trains input data through competitive learning, so that similarities in the input data are reflected in the output space, usually a two-dimensional grid. Its workflow is as follows:

[0037] Initialization phase: randomly select weight vector Wi∈R m , assume that the input data has m dimensions, and the weight vector Wi of the i-th node has m dimensions.

[0038] Competition phase: Each iteration randomly selects an input sample X∈R from the training data m , calculate the Euclidean distance between each node and the input sample, the calculation formula is as follows:

[0039]

[0040] Where d(i,X) is the Euclidean distance between node i and input sample X, and w ij is the jth weight component of the i-th node, m is the dimension, and the node with the smallest distance is selected as the best matching unit (BMU) according to the above formula:

[0041]

[0042] Update phase: Update the weights of the BMU and its neighborhood nodes to shorten the gap with the input sample. The formula is as follows:

[0043] W i (t+1)=W i (t)+α(t)·h i,j (t)·(XW i (t))

[0044] Among them, W i (t) is the weight vector of the i-th node at the t-th iteration, h i,j (t) is the neighborhood function, which represents the similarity between node i and BMU and is defined using Gaussian function:

[0045]

[0046] Where, is the distance between node i and BMU, σ(t) is the neighborhood range, and the calculation formula is shown in the formula. Its value increases with the increase of the number of iterations:

[0047]

[0048] Where σ0 is the initial neighborhood width and T is the total number of iterations.

[0049] σ(t) is the learning rate, which usually decays gradually over time t, and the decay method is:

[0050]

[0051] Where α0 is the initial learning rate and T is the total number of iterations.

[0052] Furthermore, through the updating step, the SOM network gradually adjusts the weights of the nodes so that similar data are mapped to adjacent nodes. The SOM performs similarity clustering on the 14 parameters extracted from the electrochemical experiment, finds data points with similar characteristics, and maps them onto a two-dimensional grid.

[0053] Furthermore, in step 402, after the SOM performs similarity clustering on the 14 parameters, the genetic algorithm explores different combinations of the 14 features to find the best combination that can predict the remaining service life of the battery. Ultimately, five features related to the health status of retired lead-acid batteries are determined, including three EIS features: fs2, n2, n_w and two HPPC features Rin and P.

[0054] Furthermore, in step 403, the features optimized by the genetic algorithm are input into XGBoost for training, and the XGBoost regression capability is used to estimate the remaining service life of retired lead-acid batteries. The specific process is as follows: data preprocessing; SOM feature extraction: using SOM to extract low-dimensional features; feature selection: using genetic algorithm to optimize feature selection, and selecting the optimal feature subset from the low-dimensional features output by SOM; training XGBoost model: using the optimized features to train the XGBoost model; evaluation and prediction: using the test set to evaluate the performance of the trained XGBoost model, predicting the test set, and obtaining the remaining service life of the prediction set.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] Health status estimation of retired lead-acid batteries based on the SOM-GA-XGBoost hybrid model. SOM is used to reduce the dimension of battery performance data and extract features, retaining key patterns and trends in the data and providing more efficient and representative features for subsequent modeling.

[0057] A genetic algorithm is used to further optimize the features extracted by the SOM and select the most representative and predictive feature combination. XGBoost is used to train the health status prediction model for retired lead-acid batteries, and the advantages of the gradient boosting tree are combined to achieve better performance when processing nonlinear relationships and complex data.

[0058] The characteristic parameters of the present invention are selected based on the working principle of the battery. Electrochemical tests such as EIS and HPPC are performed on retired lead-acid batteries to obtain the original parameter set characterizing the battery performance. The SOM-GA-XGBoost hybrid model is established to estimate the health status of lead-acid batteries. It can be seen that compared with the single GBDT model, the SOM-GA-XGBoost model captures trends more accurately, the changes in the predicted values ​​are basically consistent with the true values, and it can reflect the local fluctuations of the true values ​​more carefully. Through the pattern extraction of SOM and the feature optimization of GA, the hybrid model performs more robustly in areas with large local fluctuations. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is a flow chart of an embodiment of a method for estimating the state of health of retired lead-acid batteries based on a SOM-GA-XGBoost hybrid model;

[0060] Figure 2 It is a typical electrochemical impedance spectroscopy curve;

[0061] Figure 3 It is a typical voltage response in HPPC testing;

[0062] Figure 4 is an equivalent circuit diagram established according to the shape of the electrochemical impedance spectroscopy in one embodiment;

[0063] Figure 5 This is a diagram showing the electrochemical impedance spectroscopy fitting results using an equivalent circuit in one embodiment;

[0064] Figure 6 is the equivalent circuit parameter of the lead-acid battery;

[0065] Figure 7 It is the SOM similarity clustering grid diagram;

[0066] Figure 8 It is the genetic algorithm GA workflow diagram;

[0067] Figure 9 An embodiment of the invention is based on the health status estimation result of the SOM-GA-XGBoost hybrid model;

[0068] Figure 10 It is a pair of health status estimation results based on a single GBDT model. Specific implementation methods

[0069] The technical solutions in the embodiments of the present invention will be fully described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0070] See also Figure 1 The present invention proposes a health status estimation method for retired lead-acid batteries based on the SOM-GA-XGBoost hybrid model. The process is as follows: Figure 1 As shown, the specific steps include:

[0071] Step 1: Use an electrochemical workstation to measure the electrochemical impedance spectra of 104 retired lead-acid batteries.

[0072] In the present invention, the test object is a "Shuangdeng" GFM series valve-regulated sealed lead-acid battery, which has a nominal voltage of 2V, a nominal capacity of 500Ah, a charging current of 50A, a float charge voltage of 2.23V, and an equalization charge voltage of 2.3V.

[0073] In the present invention, the electrochemical impedance spectroscopy experiment uses an electrochemical workstation, and the impedance test frequency range is selected to be 0.01Hz-1kHz. The principle of measuring electrochemical impedance spectroscopy by the electrochemical workstation is as follows: a small current signal is injected into the battery, the current excitation signal and the voltage response signal are collected, and fast Fourier transform is performed on each signal. The specific calculation formula is as follows:

[0074]

[0075] Where i[n] is the current excitation signal, u[n] is the voltage response signal, ω represents the angular frequency, n represents the number of discrete signal points, and j is the imaginary unit;

[0076] Calculate the impedance Z(jω) of the battery under test based on the frequency domain signals I(jω) and U(jω):

[0077] Z(jω)=U(jω) / I(jω)=Z′+Z″

[0078] Wherein, I(jω) and U(jω) are the frequency domain signals calculated by the above formula, and Z′ and Z″ represent the real impedance and imaginary impedance of the battery to be tested, respectively.

[0079] In the present invention, a typical electrochemical impedance spectroscopy is as follows Figure 2 As shown in the figure, the electrochemical impedance spectrum can be divided into three parts according to the frequency range: high frequency part, medium frequency part and low frequency part. Each part has a corresponding circuit element description, such as Figure 4The figure shows the established equivalent circuit, in which the high-frequency response is described by the inductor L, the ohmic region is described by the resistor, the intermediate frequency part is described by two constant phase elements, the SEI film impedance RSEI and the charge transfer impedance Rct, and the low-frequency part is described by the Warburg element Rw.

[0080] Step 2: Perform a hybrid pulse power characteristic test on 104 retired lead-acid batteries in a charge and discharge tester.

[0081] In the present invention, the specific steps of the hybrid pulse power characteristic test experiment are as follows: the retired lead-acid battery is left to stand for thirty minutes; the battery after standing is charged and discharged in a charge and discharge tester at a 0.1C pulse, one pulse 10s; the battery is charged and discharged in a charge and discharge tester at a 0.2C pulse, one pulse 10s, a typical voltage response is as follows Figure 3 shown.

[0082] Step 3: Fit the electrochemical impedance spectra of 104 retired lead-acid batteries from the power grid, analyze the hybrid pulse power characteristic test data, and obtain performance parameters that characterize the internal state of retired lead-acid batteries from the power grid.

[0083] In the present invention, the particle swarm optimization (PSO) algorithm is used to automatically fit the constructed Randles equivalent circuit model, and the fitting effect is as follows: Figure 5 As shown, the fitted chi-square value is 1.2983e-06, as Figure 6 Shown are the 11 equivalent circuit parameters EIS_fs = {R0, L1, R1, fs1, n1, R2, fs2, n2, Rw, tau_w, n_w} obtained by fitting.

[0084] In this embodiment, three parameters closely related to the battery health status are extracted based on the HPPC experiment: HPPC_fs = {Vdrop, Rin, P}:

[0085] Battery voltage response: This type of data can reveal the changes in the battery's internal resistance and the battery's charge and discharge capabilities. The formula for calculating the voltage drop after the battery experiences a high-frequency pulse response is as follows:

[0086] V drop =V max -V min

[0087] Where V max Indicates the maximum voltage across the battery in a high-frequency pulse response, V min Indicates the minimum voltage across the battery within a high-frequency pulse response.

[0088] Internal resistance: This parameter is one of the important indicators of battery degradation. Its value increases as the battery ages. The calculation formula is as follows:

[0089]

[0090] Where ΔV represents the change in voltage across the battery during a high-frequency pulse response, and ΔI represents the change in current inside the battery during a high-frequency pulse response.

[0091] Power response: During a high-frequency pulse response, the output power of the battery is calculated as follows:

[0092] P(t)=V(t)×I(t)

[0093] Where V(t) represents the voltage response during a high-frequency pulse, and I(t) represents the current response during a high-frequency pulse.

[0094] Step 4: Establish a SOM-GA-XGBoost hybrid model to determine the characteristic parameters related to the remaining health state of the battery and realize the health state estimation of retired lead-acid batteries.

[0095] In the present invention, the health status estimation of retired lead-acid batteries based on the SOM-GA-XGBoost hybrid model requires three steps:

[0096] Step 401: using a self-organizing map unsupervised learning method to perform similarity clustering on the parameters characterizing battery performance extracted from the experiment, find data points with similar characteristics, and map them onto a two-dimensional grid;

[0097] Step 402: After the SOM performs similarity clustering on the parameters, a genetic algorithm is used to perform heuristic optimization to explore different combinations of performance parameters and find the best combination solution that can predict the remaining useful life (RUL) of the battery;

[0098] Step 403: Input the features optimized by the genetic algorithm into XGBoost for training, and use the powerful regression ability of XGBoost to estimate the remaining useful life (RUL) of retired lead-acid batteries.

[0099] In the present invention, the self-organizing map unsupervised learning method in step 401 maps high-dimensional data to a low-dimensional space while maintaining the topological structure of the data. Unlike traditional neural networks, SOM trains input data through competitive learning, so that the similarity of the input data is reflected in the output space (usually a two-dimensional grid). Its workflow is as follows:

[0100] Initialization phase: randomly select weight vector Wi∈R m , assume that the input data has m dimensions, and the weight vector Wi of the i-th node has m dimensions.

[0101] Competition phase: Each iteration randomly selects an input sample X∈R from the training data m , calculate the Euclidean distance between each node and the input sample, the calculation formula is as follows:

[0102]

[0103] Where d(i,X) is the Euclidean distance between node i and input sample X, and w ij is the jth weight component of the i-th node, m is the dimension, and the node with the smallest distance is selected as the best matching unit (BMU) according to the above formula:

[0104]

[0105] Update phase: Update the weights of the BMU and its neighborhood nodes to shorten the gap with the input sample. The formula is as follows:

[0106] W i (t+1)=W i (t)+α(t)·h i,j (t)·(XW i (t))

[0107] Among them, W i (t) is the weight vector of the i-th node at the t-th iteration, h i,j (t) is the neighborhood function, which represents the similarity between node i and BMU and is defined using Gaussian function:

[0108]

[0109] Where, is the distance between node i and BMU, σ(t) is the neighborhood range, and the calculation formula is shown in the formula. Its value increases with the increase of the number of iterations:

[0110]

[0111] Where σ0 is the initial neighborhood width and T is the total number of iterations.

[0112] σ(t) is the learning rate, which usually decays gradually over time t, and the decay method is:

[0113]

[0114] Where α0 is the initial learning rate and T is the total number of iterations.

[0115] Furthermore, through the updating step, the SOM network gradually adjusts the weights of the nodes so that similar data are mapped to adjacent nodes. The SOM performs similarity clustering on the 14 parameters extracted from the electrochemical experiment, finds data points with similar characteristics, and maps them onto a two-dimensional grid.

[0116] In the present invention, after SOM performs similarity clustering on 17 parameters, step 402 uses genetic algorithm to explore different combinations of 14 features to find the best combination solution that can predict the remaining useful life (RUL) of the battery. The workflow is as follows: Figure 9 As shown in the figure, five features related to the health status of retired lead-acid batteries were finally identified, including three EIS features: fs2, n2, n_w and two HPPC features Rin and P.

[0117] In the present invention, step 403 inputs the features optimized by the genetic algorithm into XGBoost for training, and uses the powerful regression ability of XGBoost to estimate the remaining useful life (RUL) of retired lead-acid batteries.

[0118] In this embodiment, the results of health status estimation of retired lead-acid batteries based on the SOM-GA-XGBoost hybrid model are as follows: Figure 9 As shown in the figure, the SOM-GA-XGBoost hybrid model has a strong trend-capturing capability. The trend of the predicted values ​​is generally consistent with the true values, indicating that the model can effectively capture the dynamic changes in the health status (RUL) of lead-acid batteries. Even if there are errors in the specific predicted values ​​of some samples, the overall prediction trend of the model is consistent with the true value. This is very important for practical battery health management. When it comes to the screening and reorganization of retired lead-acid batteries, trends can better guide decisions than specific values. The health status of lead-acid batteries exhibits a complex nonlinear relationship with their internal resistance, voltage, temperature, and other characteristics. Traditional linear models are difficult to accurately model. However, XGBoost, as a gradient boosting tree model, can effectively capture these complex nonlinear characteristics.

[0119] A comparative example is also included, using a single GBDT model to estimate the health status of retired lead-acid batteries. The workflow is as follows:

[0120] Step 1: Input the feature matrices EIS_fs and HPPC_fs, where EIS_fs = {R0, L1, R1, fs1, n1, R2, fs2, n2, Rw, tau_w, n_w}; HPPC_fs = {Vdrop, Rin, P} and the target variable RUL.

[0121] Step 2: Use the training set to fit the GBDT model. During the training process, the nonlinear relationship between the feature parameters and the remaining health (RUL) of the battery is automatically learned.

[0122] Step 3: Use grid search to optimize hyperparameters, evaluate the effects of different parameter combinations, select the parameters with the best performance, and use the test set data to verify the model prediction performance;

[0123] Step 4: Compare the predicted SOH with the measured SOH to assess the model's error margin.

[0124] like Figure 10 The figure shows the results of using a single GBDT model to estimate the health status of retired lead-acid batteries. As can be seen from the figure, the model directly uses the original features for training, and none of the features undergo feature dimensionality reduction and optimization. Due to the insufficient contribution of some features to the model, the model performance is affected. When dealing with complex nonlinear relationships, the single GBDT model suffers from a decrease in robustness due to the redundant number of features. The hybrid model extracts key features through the SOM and removes redundant and invalid features through the GA. This makes the model more resistant to data noise, the error distribution of the prediction results is more uniform, and it exhibits high robustness, maintaining stable performance even on the test set.

[0125] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method for estimating the health status of retired lead-acid batteries based on a SOM-GA-XGBoost hybrid model, characterized in that: The following steps are involved: S1: Measure the electrochemical impedance spectroscopy of retired lead-acid batteries; S2: Conduct hybrid pulse power characteristic test on retired lead-acid batteries; S3: Fit the electrochemical impedance spectrum of retired lead-acid batteries to obtain an equivalent circuit parameter set that characterizes the internal performance characteristics of retired lead-acid batteries; analyze the mixed pulse power characteristic test data to obtain a performance parameter set that characterizes the internal state of retired lead-acid batteries; S4: Establish a SOM-GA-XGBoost hybrid model to determine the characteristic parameters related to the remaining health state of the battery and realize the health state estimation of retired lead-acid batteries.

2. The method for estimating the health status of retired lead-acid batteries based on the SOM-GA-XGBoost hybrid model according to claim 1 is characterized in that: Step S1 measures the electrochemical impedance spectrum of the retired lead-acid battery by using an electrochemical workstation, injecting a small current signal into the battery, collecting the current excitation signal i[n] and the voltage response signal u[n], and performing fast Fourier transform (FFT) on them to obtain frequency domain signals I(jω) and U(jω): Where i[n] is the current excitation signal, u[n] is the voltage response signal, ω represents the angular frequency, n represents the number of discrete signal points, and j is the imaginary unit; Calculate the impedance Z(jω) of the battery under test based on the frequency domain signals I(jω) and U(jω): Z(jω)=U(jω) / I(jω)=Z′+Z″ Wherein, I(jω) and U(jω) are the frequency domain signals calculated by the above formula, and Z′ and Z″ represent the real impedance and imaginary impedance of the battery to be tested, respectively.

3. The method for estimating the health status of retired lead-acid batteries based on the SOM-GA-XGBoost hybrid model according to claim 1, characterized in that: Step S2 is to perform a mixed pulse power characteristic test, specifically, after the retired lead-acid battery is left to stand for thirty minutes, pulses with a length of 10 seconds and amplitudes of 0.1C and 0.2C are intermittently injected.

4. The method for estimating the health status of retired lead-acid batteries based on the SOM-GA-XGBoost hybrid model according to claim 1, characterized in that: Step S3 of fitting the electrochemical impedance spectrum of the retired lead-acid battery specifically involves establishing an equivalent circuit model based on the measured electrochemical impedance spectrum curve, fitting the electrochemical impedance spectrum of the retired lead-acid battery, and thereby obtaining an equivalent circuit parameter set that characterizes the internal performance characteristics of the retired lead-acid battery.

5. The method for estimating the health status of retired lead-acid batteries based on the SOM-GA-XGBoost hybrid model according to claim 4 is characterized in that: The equivalent circuit model is: L1R0(Q1R1)(Q2R2)Rw, where the high-frequency part is described by inductance L, the ohmic region is described by resistance, the medium-frequency part is described by two constant-phase elements, SEI film impedance RSEI and charge transfer impedance Rct, and the low-frequency part is described by Warburg element Rw.

6. The method for estimating the health status of retired lead-acid batteries based on the SOM-GA-XGBoost hybrid model according to claim 1, characterized in that: The step S3 analyzes the mixed pulse power characteristic test data, obtains a performance parameter set characterizing the internal state of the retired lead-acid battery, and calculates the features in the performance parameter set.

7. The method for estimating the health status of retired lead-acid batteries based on the SOM-GA-XGBoost hybrid model according to claim 1, characterized in that: Step S4 establishes a SOM-GA-XGBoost hybrid model to determine characteristic parameters related to the remaining state of health of the battery and estimate the state of health of retired lead-acid batteries. Specifically, the following steps are included: S401: Using a self-organizing map (SOM) unsupervised learning method, the equivalent circuit parameter set and the performance parameter set extracted in step S3 are clustered based on similarity, data points with similar characteristics are found, and mapped onto a two-dimensional grid; S402: After similarity clustering is performed on the parameter set, a genetic algorithm is used to perform heuristic optimization to explore different combinations of performance parameters and find the best combination solution that can predict the remaining useful life (RUL) of the battery; S403: Input the features optimized by the genetic algorithm into XGBoost for training, and use the regression ability of XGBoost to estimate the remaining useful life (RUL) of retired lead-acid batteries.

8. The method for estimating the health status of retired lead-acid batteries based on the SOM-GA-XGBoost hybrid model according to claim 6, characterized in that: In step S401, SOM trains the input data by competitive learning, and the similarity of the input data is reflected in the output space. The specific working process is as follows: In the initialization phase, a weight vector Wi∈R is randomly selected m , set the input data to be m-dimensional, and the weight vector Wi of the i-th node has dimension m; In the competition phase, an input sample X∈R is randomly selected from the training data in each iteration. m , calculate the Euclidean distance between each node and the input sample, and select the node with the smallest distance as the best matching unit (BMU); In the update phase, the weights of the best matching unit and the nodes in its neighborhood are updated to shorten the gap with the input sample. Through the update step, the SOM network gradually adjusts the weights of the nodes so that similar data are mapped to adjacent nodes. The equivalent circuit parameter set and the performance parameter set extracted in step S3 are clustered by similarity to find data points with similar characteristics and map them to a two-dimensional grid.

9. The method for estimating the health status of retired lead-acid batteries based on the SOM-GA-XGBoost hybrid model according to claim 8, characterized in that: In step S402, a heuristic algorithm (GA) is used when optimizing feature selection or adjusting hyperparameters. The specific process is as follows: a genetic algorithm is used to randomly generate a population and calculate the RMSE and MAE of the training results; the initial individuals for the next training are selected based on fitness to generate new individuals; diversity is introduced and more search space is explored; after reaching a preset threshold, the search is stopped and the optimal solution is returned to obtain the best feature combination solution.

10. The method for estimating the health status of retired lead-acid batteries based on the SOM-GA-XGBoost hybrid model according to claim 8, characterized in that: In step S403, the feature combination optimized by the genetic algorithm is input into XGBoost for training, and the trained XGBoost model is evaluated using a test set to estimate the remaining useful life (RUL) of retired lead-acid batteries.

Images