Lithium battery state of health estimation method considering inconsistency under limited monitoring

Abstract

The present application relates to the technical field of lithium ion battery health monitoring, in particular to a lithium battery health state estimation method considering inconsistency under limited monitoring, comprising: S1: constructing battery groups with different inconsistency degrees; S2: conducting battery group aging experiments on each group of battery groups to obtain operation data and corresponding battery group SOH true values of each group of battery groups; S3: analyzing to obtain health characteristics of the monitored single battery and estimating health characteristics of the unmonitored single battery; S4: calculating aging characteristics and inconsistency characteristics of the battery group and constructing a battery group optimal feature subset; S5: training a SOH prediction model through the battery group optimal feature subset and the corresponding battery group SOH true values; S6: obtaining operation data of a to-be-tested battery group and inputting the to-be-tested battery group optimal feature subset into the trained SOH prediction model to output a battery group SOH prediction value. The present application realizes accurate estimation of lithium ion battery group SOH with different consistency degrees.

Description

Technical Field

[0001] This invention relates to the field of lithium-ion battery health monitoring technology, and specifically to a method for estimating the health status of lithium batteries that takes into account inconsistencies under limited monitoring conditions. Background Technology

[0002] With the widespread application of lithium-ion batteries in energy storage, electric vehicles, and various electrical equipment, it is typically necessary to connect dozens to thousands of individual cells in series, parallel, or hybrid topologies to form battery packs. However, during long-term charge-discharge cycles, individual cells undergo irreversible physicochemical changes such as the growth of a solid electrolyte interface film, loss of active lithium, and degradation of active materials, leading to a decrease in capacity, an increase in internal resistance, and a reduction in power output, thereby affecting the performance and reliability of the entire battery pack.

[0003] Meanwhile, due to factors such as production and assembly processes, storage environment, differences in self-discharge, and operating conditions, there are inherent and acquired inconsistencies among the individual cells within a battery pack. These inconsistencies intensify with aging cycles, causing some cells to reach their charge / discharge cutoff voltage first, potentially leading to overcharging, over-discharging, thermal runaway, and short circuits. For series-connected battery packs, the maximum usable capacity is typically limited by the cell with the highest aging rate and smallest capacity; therefore, cell inconsistency is a significant factor affecting the accuracy of SOH (State of Harm) estimation for battery packs.

[0004] Existing battery pack SOH estimation methods mainly fall into two categories: one treats the battery pack as a "large battery" and directly applies the SOH estimation model of a single cell; the other considers the inconsistencies between cells within the pack to construct a battery pack SOH estimation model. While the former method is relatively simple to implement, it cannot effectively characterize the differences between cells within the pack and their evolution process; the latter method, although able to account for inconsistencies, typically relies on comprehensive parameter monitoring of all cells and the deployment of numerous sensors, making it technically and economically infeasible for large-scale battery systems.

[0005] Furthermore, in practical engineering, due to cost constraints, sensor failures, communication interruptions, or limited deployment conditions, battery packs often only monitor a portion of individual cells, resulting in incomplete data. Existing methods for scenarios with incomplete data mostly rely on data augmentation or missing data completion, but due to the lack of electrochemical mechanism constraints, they are difficult to accurately reflect the actual capacity deviations between cells during the full aging stage of the battery pack, thus affecting the accuracy of SOH estimation.

[0006] In summary, it is urgent to further clarify the impact mechanism of intra-group inconsistency on the battery pack aging process, and on this basis, to construct a battery pack SOH estimation model that is suitable for incomplete data monitoring conditions and takes into account inconsistency characteristics, so as to achieve accurate assessment of the battery pack health status. Summary of the Invention

[0007] To address the shortcomings of the existing technologies, the technical problem to be solved by this invention is: how to provide a method for estimating the state of health (SOH) of lithium-ion battery packs that takes into account inconsistencies under limited monitoring conditions. Under limited monitoring conditions, this method combines feature extraction of monitorable individual cells, feature estimation of unmonitored individual cells, and construction of aging and inconsistency features within the pack to achieve accurate estimation of SOH for lithium-ion battery packs with different levels of consistency. This method is suitable for assessing the state of health of large battery packs in scenarios where data monitoring is incomplete.

[0008] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0009] Methods for estimating the state of health of lithium batteries that take into account inconsistencies under limited monitoring include:

[0010] S1: Construct several battery packs consisting of multiple individual cells connected in series and having different degrees of inconsistency;

[0011] S2: Conduct battery aging tests on each battery pack to obtain the operating data of each battery pack and the corresponding actual SOH value of the battery pack; the operating data of the battery pack includes the voltage data of the monitored individual cells;

[0012] S3: Perform correlation analysis on the voltage data of the monitored individual cells in each battery pack to obtain the health characteristics of the monitored individual cells, and estimate the health characteristics of the unmonitored individual cells in each battery pack based on the health characteristics of the monitored individual cells to obtain the complete health characteristic set of each battery pack.

[0013] S4: Calculate aging features and inconsistency features based on the complete health feature set of each battery pack, and sort and filter the features by importance to generate the optimal feature subset of the battery pack; the index of the optimal feature is determined in the optimal feature subset of the battery pack.

[0014] S5: Use the corresponding optimal feature subset of the battery pack and the true SOH value of the battery pack as a set of training samples; use the optimal feature subset of the battery pack corresponding to each training sample as the model input, and use the corresponding true SOH value of the battery pack as the supervision label to train the battery pack SOH regression prediction model.

[0015] S6: Obtain the operating data of the battery pack under test, calculate the health characteristics of the monitored individual cells and estimate the health characteristics of the unmonitored individual cells to obtain a complete health feature set; calculate aging characteristics and inconsistency characteristics based on the complete health feature set of the battery pack under test, and select the optimal feature based on the index of the optimal feature to generate the optimal feature subset; input the optimal feature subset of the battery pack under test into the trained SOH prediction model, and output the SOH prediction value of the battery pack under test.

[0016] Preferably, in step S1, there is a preset capacity difference between individual cells in each battery pack, so that each battery pack exhibits a different degree of inconsistency.

[0017] Preferably, in step S2, the formula for calculating the true SOH value of the battery pack is:

[0018] ;

[0019] In the formula: This represents the actual SOH value of the battery pack. This represents the maximum discharge capacity of the battery pack. For the first in the battery pack The current discharge capacity of each individual cell; This refers to the rated capacity of the battery pack.

[0020] Preferably, step S3 specifically includes the following processing steps:

[0021] S301: Extract candidate health features reflecting the aging state of the battery based on the voltage data of each monitored individual cell in the battery pack.

[0022] S302: Calculate the Pearson correlation coefficient for the same candidate health characteristics among different monitored individual cells;

[0023] S303: Select candidate health features as the health features of the monitored single cell based on the Pearson correlation coefficient;

[0024] S304: Input the health features of the monitored individual cells into the trained individual cell health feature estimation model, and output the health features of the unmonitored individual cells.

[0025] S305: Combine the health characteristics of the monitored individual cells and the estimated health characteristics of the unmonitored individual cells in order of cell number to obtain the complete health characteristic set of the battery pack.

[0026] Preferably, in step S301, the candidate health characteristics reflecting the battery aging state include any one or more of the following:

[0027] (1) Open-circuit voltage after constant current discharge;

[0028] (2) The ratio of the average charging voltage to the charging capacity;

[0029] (3) Area of ​​the voltage curve during constant current charging;

[0030] (4) Area of ​​the voltage curve during constant voltage charging;

[0031] (5) Charging cut-off voltage;

[0032] (6) Charging start voltage.

[0033] Preferably, step S4 specifically includes the following processing steps:

[0034] S401: Calculate the aging characteristics and inconsistency characteristics of all battery packs to form an initial feature set, and associate each feature in the initial feature set with the actual SOH value of the corresponding battery pack as a sample to construct the original sample set;

[0035] S402: A predetermined number of samples are randomly selected from the original sample set using the bootstrap sampling method to construct a decision tree, and the remaining samples are used as the out-of-bag data of the decision tree.

[0036] S403: The SOH prediction value of the battery pack is obtained by using the decision tree to predict the SOH of the data outside the bag, and the mean square error between the predicted SOH value and the actual SOH value of the battery pack is calculated as the original outside bag error.

[0037] S404: Randomly permuting all values ​​of a specific feature in the out-of-bag data of the decision tree to achieve out-of-bag data perturbation; using the decision tree to re-predict the SOH of the perturbed out-of-bag data to obtain a new SOH prediction value for the battery pack, and calculating the mean square error between the new SOH prediction value and the actual SOH as the perturbation out-of-bag error;

[0038] S405: Calculate the difference between the perturbation out-of-bag error and the original out-of-bag error as the difference of this specific feature in the corresponding decision tree;

[0039] S406: Repeat steps S402 to S405 to calculate the difference of the specific feature across all decision trees, and take the average of the differences of the specific feature across all decision trees as the importance score of the specific feature;

[0040] S407: Repeat steps S402 to S406 to obtain the importance scores of all features, and sort the importance scores of all features in descending order from high to low.

[0041] S408: Select the previous one in a step-by-step incremental manner The model is trained using 10 features as input features, and the accuracy of the model is recorded to generate a curve reflecting the change in the accuracy of the model with the number of input features.

[0042] S409: Identify the performance inflection point of the curve and select the number of features corresponding to that inflection point. The top-ranked candidates with the highest importance scores These features are selected as the optimal feature subset for the battery pack.

[0043] Preferably, in step S401, the formula for calculating the aging characteristics of the battery pack is:

[0044] ;

[0045] In the formula: For the battery pack A characteristic of aging, Indicates the battery pack hierarchy. Indicates the type of health characteristic; To indicate the first The first single cell One health characteristic; Indicates the number of individual cells in the battery pack; where the battery pack has A characteristic of aging.

[0046] Preferably, in step S401, the formula for calculating the inconsistency characteristics of the battery pack is:

[0047] ;

[0048] In the formula: Indicates the first The first single cell The health characteristics and all individual battery cells The difference between the average values ​​of the first health characteristic, i.e., the first health characteristic of the battery pack. One inconsistent feature; Indicates the number of individual cells in the battery pack; where the battery pack has A characteristic of aging.

[0049] Preferably, in step S5, the SOH prediction model is the SVM-Adaboost ensemble model;

[0050] The steps involved in training the SVM-Adaboost ensemble model include:

[0051] S501: Acquisition Each training sample set includes the optimal feature subset of the battery pack and the corresponding true SOH value of the battery pack.

[0052] S502: Random selection The training dataset is composed of groups of training samples, and the sample weights of each group of training samples are set to the same initial value;

[0053] S503: The Adaboost model's iterative loop is entered based on the current round's sample weight distribution, the optimal feature subset of the battery pack, and the battery pack's true SOH value.

[0054] S5031: In the In each iteration, a weak support vector machine learner is trained using the current sample weight distribution;

[0055] S5032: Use a support vector machine weak learner to predict the training dataset and obtain the battery pack SOH prediction sequence. And calculate the weighted prediction error of the weak learner in the support vector machine. ;

[0056] The formula is expressed as:

[0057] ;

[0058] In the formula: For the first In the first iteration The sample weights of each training sample;

[0059] S5033: Prediction error based on the weak learner of the support vector machine Calculate its model weights in the final strong learner. ;

[0060] The formula is expressed as:

[0061] ;

[0062] S5034: Update the sample weights of each training sample in the training dataset to obtain the next iteration. The sample weight distribution is determined, and the process returns to step S5031;

[0063] The formula is expressed as:

[0064] ;

[0065] In the formula: For the first In the first iteration The sample weights of each training sample; For the first In the first iteration The sample weights of each training sample; This is a normalization factor used to normalize the updated sample weights. For the support vector machine weak learner in the th Model weights in each iteration; For the first training samples The actual SOH value of the battery pack; For the first Round of iteration training samples The sequence of predicted SOH values ​​for the battery pack;

[0066] S5035: Repeat steps S5031 to S5034 to obtain A trained support vector machine weak learner, and the model weights corresponding to each support vector machine weak learner;

[0067] S504: Based on The model weights of the support vector machine weak learners are paired. Linear fusion of multiple weak support vector machine learners yields a linear combination of multiple weak classifiers. ;

[0068] The formula is expressed as:

[0069] ;

[0070] In the formula: For the first A support vector machine weak learner is used to learn any input feature vector. The predicted value;

[0071] S505: Based on linear combination of multiple weak classifiers Constructing a strong prediction function As a pre-trained SVM-Adaboost model;

[0072] The formula is expressed as:

[0073] ;

[0074] In the formula: This is a sign function. When the input is greater than 0, the output is 1; when the input is less than 0, the output is -1; and when the input is equal to 0, the output is 0.

[0075] Preferably, in step S5, the performance of the SVM-Adaboost ensemble model is evaluated using mean absolute error, root mean square error, and mean absolute percentage error.

[0076] The calculation formula is:

[0077] ;

[0078] ;

[0079] ;

[0080] In the formula: , , These are mean absolute error, root mean square error, and mean absolute percentage error, respectively. This is the predicted SOH value for the battery pack; This represents the actual SOH value of the battery pack. This represents the number of training samples.

[0081] The lithium battery health status estimation method for limited monitoring and inconsistency assessment in this invention has the following advantages compared with existing technologies:

[0082] This invention enables the estimation of the health characteristics of unmonitored cells using the health characteristics of monitorable cells without requiring comprehensive monitoring of all individual cells. Furthermore, it constructs a feature system that simultaneously characterizes the aging degree of the battery pack (i.e., intra-pack aging characteristics) and the level of inconsistency within the pack (i.e., inconsistency characteristics). By combining feature importance ranking and filtering using a random forest algorithm, the optimal features for the battery pack are selected, achieving accurate estimation of the state of health (SOH) of lithium-ion battery packs with varying degrees of consistency. This significantly improves the accuracy and robustness of SOH estimation and is suitable for assessing the health status of large battery packs in scenarios with incomplete data monitoring. Moreover, compared to existing methods that treat the battery pack as a "large battery" or rely on the deployment of all sensors, this invention balances engineering feasibility and techno-economic efficiency. It not only effectively reduces sensor configuration costs and monitoring complexity but also provides a reference for optimizing the number and placement of sensors in large battery packs. Attached Figure Description

[0083] To make the objectives, technical solutions, and advantages of the invention clearer, the invention will now be described in further detail with reference to the accompanying drawings, wherein:

[0084] Figure 1 This is a logic block diagram for a lithium battery health state estimation method that takes into account inconsistencies under limited monitoring.

[0085] Figure 2 SOH estimation results for the battery pack: Figure 2 (a) Figure 2 (b) Figure 2 (c) Figure 2 (d) are the SOH estimation results for aging cycles G1, G2, G3 and G4, respectively. Detailed Implementation

[0086] 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 with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but only to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0087] The following detailed explanation illustrates the specific implementation methods:

[0088] Example:

[0089] This embodiment discloses a method for estimating the health status of lithium batteries under limited monitoring and considering inconsistencies.

[0090] like Figure 1 As shown, a method for estimating the state of health of lithium batteries considering inconsistencies under limited monitoring includes:

[0091] S1: Construct several battery packs consisting of multiple individual cells connected in series and having different degrees of inconsistency;

[0092] S2: (Through the established battery pack aging test platform) Conduct battery pack aging tests on each group of battery packs to obtain the operating data of each group of battery packs and the corresponding actual SOH value of the battery packs; the operating data of the battery packs includes the voltage data of the monitored individual cells, the total voltage data of the battery pack, and the current data of the battery pack; among them, the total voltage and current data of the battery pack are mainly used to characterize the overall charging and discharging process of the battery pack and to calculate the battery pack capacity and the actual SOH value. In the process of health feature extraction and health feature estimation of unmonitored cells, the voltage data of the monitored individual cells are mainly used.

[0093] S3: Perform correlation analysis on the voltage data of the monitored individual cells in each battery pack to obtain the health characteristics of the monitored individual cells, and estimate the health characteristics of the unmonitored individual cells in each battery pack based on the health characteristics of the monitored individual cells to obtain the complete health characteristic set of each battery pack.

[0094] S4: Based on the complete health feature set of each battery pack, calculate aging features and inconsistency features, and use the random forest algorithm to sort and filter the features by importance to generate the optimal feature subset of the battery pack; the index of the optimal feature is determined in the optimal feature subset of the battery pack;

[0095] S5: Use the corresponding optimal feature subset of the battery pack and the true SOH value of the battery pack as a set of training samples; use the optimal feature subset of the battery pack corresponding to each training sample as the model input, and use the corresponding true SOH value of the battery pack as the supervision label to train the battery pack SOH regression prediction model.

[0096] S6: Obtain the operating data of the battery pack under test, calculate the health characteristics of the monitored individual cells and estimate the health characteristics of the unmonitored individual cells to obtain a complete health feature set; calculate aging characteristics and inconsistency characteristics based on the complete health feature set of the battery pack under test, and select the optimal feature based on the index of the optimal feature to generate the optimal feature subset; input the optimal feature subset of the battery pack under test into the trained SOH prediction model, and output the SOH prediction value of the battery pack under test.

[0097] In this embodiment, the health features of the monitored individual cells are extracted according to the health feature extraction rules and feature construction rules determined in the training phase (i.e., steps S3 to S4), and the estimated health features of the unmonitored individual cells are obtained through the trained individual cell health feature estimation model. Based on the health features of the monitored individual cells and the estimated health features of the unmonitored individual cells, the aging features and inconsistency features of the battery pack under test are calculated, and the optimal feature subset of the battery pack under test is formed according to the optimal feature index determined in the training phase.

[0098] This invention first establishes a battery pack aging test platform and collects operational data of battery packs with different individual cell differences during the charging and discharging process. The operational data includes the voltage of individual cells, the total voltage of the battery pack, and the battery pack current. Second, by analyzing the charging voltage change trends of monitorable cells, and utilizing the strong correlation between characteristics of individual cells within the battery pack, health characteristics of individual cells are selected. Based on the extracted health characteristics of monitorable cells, the health characteristics of unmonitored cells are estimated. Then, based on the obtained individual cell characteristics, intra-group features reflecting battery pack aging and inconsistency are constructed. The random forest algorithm is used to rank and optimize the importance of these features, resulting in an optimal feature subset. Finally, an Adaboost-based ensemble support vector machine (SVM) model is established, using the optimal feature subset of the battery pack as input, to accurately estimate the state of harmonics (SOH) of battery packs with different levels of consistency.

[0099] To better illustrate the technical solution of the present invention, this embodiment will be described in more detail through the following parts.

[0100] I. Battery pack used in the experiment

[0101] In practice, each battery pack consists of four batteries connected in series. There are pre-defined capacity differences between the individual cells in each battery pack, resulting in varying degrees of inconsistency among the different battery packs.

[0102] In this embodiment, four battery packs were selected, and different degrees of inconsistency were designed based on the capacity differences of individual cells. Specifically, the first battery pack had the smallest capacity difference between individual cells, representing a battery pack with high consistency; the second battery pack had a medium capacity difference to simulate common inconsistencies in reality; the third battery pack had a large capacity difference to simulate a situation with poor battery pack inconsistency; and the fourth battery pack was set with the largest capacity difference to examine the SOH estimation performance under extreme inconsistency conditions.

[0103] II. Battery Pack Aging Test

[0104] In the specific implementation process, a battery pack aging test platform is built to collect operational data of battery packs with different individual battery cells during the charging and discharging process. The battery pack aging test platform includes a host unit, a paperless recorder, a battery testing system, and a battery pack. The Lamborghini testing system is used for charging and discharging the battery pack. The host unit implements command control of the battery tester. The paperless recorder can measure and record data such as voltage and current of each battery in the battery pack, with a sampling frequency of 1Hz, and store the data in the host unit. This invention uses NCR18650GA batteries with a rated capacity of 3350mA·h, and charging and discharging cutoff voltages of 4.2V and 2.5V, respectively. A small battery module consisting of four batteries connected in series is used for the battery pack aging experiment, and the experimental temperature is set at 25℃.

[0105] The health status of a battery pack is affected by the "weakest link effect" of individual cells within the pack, meaning the maximum discharge capacity of the entire pack is determined by the cell with the highest degree of aging. This invention focuses on series-connected battery packs. Because the capacities of individual cells vary, the cell with the highest degree of aging always reaches its charge / discharge cutoff voltage first. If the pack continues to operate at this point, it will cause overcharging and over-discharging of that cell, affecting overall safety and performance. Therefore, the usable capacity of the battery pack is primarily limited by the cell with the highest degree of aging; that is, the maximum discharge capacity of the entire pack is determined by the cell with the smallest capacity. From a capacity perspective, the formula for calculating the true SOH value of the battery pack in this invention is:

[0106] ;

[0107] In the formula: This represents the actual SOH value of the battery pack. This represents the maximum discharge capacity of the battery pack. For the first in the battery pack The current discharge capacity of each individual cell; This refers to the rated capacity of the battery pack.

[0108] The battery testing process includes module-level and cell-level testing. To ensure that the state-of-the-art (SOH) differences of individual cells within the four battery packs are 0.5%, 1%, 1.5%, and 2.5%, respectively, the individual cells need to be aged to the corresponding capacities. Before the battery pack aging test, the initial capacity of each of the four individual cells was tested separately. Then, they were connected in series to form a battery pack, and the aging cycle began. One cycle of the battery pack consisted of a constant-current, constant-voltage full charge followed by a constant-current discharge. The capacity of each individual cell was tested every three cycles, and electrochemical impedance spectroscopy (EIS) tests were performed on the individual cells on an electrochemical workstation, with a frequency range of 0.01 Hz to 10 kHz, until the battery pack capacity was less than 80% of its initial capacity. Specifically, before each reassembly of the battery pack, the initial charge of each individual cell was recharged to avoid affecting the inconsistencies between the cells within the battery pack.

[0109] III. Screening and Estimation of Health Characteristics

[0110] In the specific implementation process, to investigate the impact of incomplete battery pack data on SOH estimation, this invention uses a battery pack consisting of four individual cells connected in series and designs a series of experiments. In the experiments, the battery pack is equipped with different numbers of voltage sensors to monitor its operating status, and simultaneously equipped with a current sensor to comprehensively monitor current changes. Specifically, four different voltage sensor configurations were set up: monitoring the voltage of only one individual cell, monitoring the voltage of two individual cells, monitoring the voltage of three individual cells, and monitoring the voltage of all four individual cells. This configuration aims to evaluate the impact of incomplete data on the accuracy of SOH estimation under different monitoring numbers.

[0111] During the experiment, the battery pack will undergo standardized charge-discharge cycle testing. This experiment aims to quantify the specific impact of the number of sensors deployed on the accuracy of SOH estimation and to explore the optimal balance between cost-effectiveness and monitoring accuracy. The experimental results will provide important reference for the design of battery pack monitoring systems.

[0112] Specifically, the processing steps for constructing a complete health feature set include:

[0113] S301: Extract candidate health features reflecting the aging state of the battery based on the voltage data of each monitored individual cell in the battery pack.

[0114] S302: Calculate the Pearson correlation coefficient for the same candidate health characteristics among different monitored individual cells;

[0115] S303: Select candidate health features as the health features of the monitored single cell based on the Pearson correlation coefficient; specifically, select candidate health features with a correlation coefficient exceeding a threshold (e.g., 0.8) as the health features of the monitored single cell.

[0116] S304: Input the health features of the monitored individual cells into the trained individual cell health feature estimation model, and output the health features of the unmonitored individual cells.

[0117] In this embodiment, the single-cell health feature estimation model is an SVM-Adaboost model. During the experimental calibration phase, a single-cell health feature estimation sample set is constructed based on the complete monitoring data of each training battery pack. The health features of the monitored single cells under the preset monitoring configuration are used as the model input, and the health features of the unmonitored single cells under the same aging cycle are used as the supervision labels to train the SVM-Adaboost model, thus obtaining the single-cell health feature estimation model.

[0118] S305: Combine the health characteristics of the monitored individual cells and the estimated health characteristics of the unmonitored individual cells in order of cell number to obtain the complete health characteristic set of the battery pack.

[0119] Since the battery pack terminal voltage is a linear superposition of the voltages of individual cells, the performance degradation of any single cell will synchronously affect the overall change pattern of the pack terminal voltage. Therefore, the individual cell voltage characteristics not only reflect the health status of the individual cell, but also directly influence the battery pack voltage characteristics through a series coupling mechanism. Thus, in battery pack SOH estimation, individual cell voltage characteristics are an important carrier for capturing inconsistent evolution and overall degradation trends. Their change mechanism can map from microscopic aging characteristics to pack-level performance degradation, making them an indispensable key information source for improving the accuracy of battery pack-level health assessment.

[0120] Specifically, candidate health characteristics reflecting battery aging include any one or more of the following:

[0121] (1) HI1: Open circuit voltage V1 after constant current discharge;

[0122] (2) HI2: The ratio of average charging voltage (voltage-time curve) to charging capacity V mean / Q charge ;

[0123] (3) HI3: Area A1 of the voltage curve during constant current charging;

[0124] (4) HI4: Area A2 of the voltage curve during constant voltage charging;

[0125] (5) HI5: Charging cutoff voltage V ce ;

[0126] (6) HI6: Charging start voltage V cs .

[0127] The candidate health characteristics identified in this invention not only reflect the health status of individual cells, but also provide a comprehensive assessment of the health status of the entire battery pack through accumulation or integration.

[0128] To estimate and predict battery pack lifespan, appropriate input features need to be selected to reflect the aging status of the battery pack. However, the trends of different HI values ​​vary, making it impossible to directly determine the correlation between HI and capacity. Therefore, this invention quantifies the correlation between HI and capacity by calculating a correlation coefficient. Specifically, the Pearson correlation coefficient is used to calculate the correlation between the features and the capacity of the four battery packs. The Pearson correlation coefficient formula is as follows:

[0129] ;

[0130] In the formula: X and Y represent the two types of sample populations, respectively; This represents the expected value of the calculated data. Pearson values ​​range from [-1, 1], with a larger absolute value indicating a higher correlation between the two classes of samples. A comparison of the correlation coefficients between various battery indirect health factors and capacity shows that HI1 and HI2 have significantly higher correlation coefficients than other features, thus they can be used to construct battery pack aging and inconsistency characteristics.

[0131] During the aging process of a battery pack, all individual cells operate in the same environment and follow the same charge-discharge control strategy. This consistent external condition means that each individual cell has similar aging drivers at a macroscopic level. Therefore, although the specific aging rate of each cell may differ due to factors such as manufacturing variations, inconsistent local thermal management, or deviations in internal electrochemical reactions, their overall degradation trend is generally highly consistent. In this context, the health characteristics of individual cells typically exhibit a strong correlation.

[0132] To verify the strong correlation between health characteristics of individual battery cells, this invention uses the aforementioned Pearson correlation coefficient to measure the correlation between characteristics of individual battery cells. In the four battery packs, HI1 and HI2 showed high correlations among different individual cells. These calculations further validate the feasibility of using characteristics of some monitorable cells to estimate characteristics of unmonitored cells.

[0133] Based on the high correlation between HI1 and HI2 among different individual cells, the HI1 and HI2 features of unmonitored cells can be accurately estimated by monitoring the HI1 and HI2 features extracted from individual cells. Specifically, by using the HI1 or HI2 features of the monitorable individual cells as input to the SVM-Adaboost model, the estimation algorithm outputs the HI1 or HI2 features of the unmonitored individual cells.

[0134] IV. Optimal Feature Subset of Battery Pack

[0135] In its implementation, the Random Forest (RF) algorithm, based on CART decision trees and the Bagging ensemble strategy, can randomly generate training subsets and out-of-bag data to construct decision trees without relying on external factors. Its core idea is to evaluate the impact of noise on the model's predictive performance, and then calculate the feature's importance score. A higher score indicates a more significant impact of the feature on the model's prediction results.

[0136] In this invention, the collected data comes from the actual operation of the battery pack under multiple aging cycles, containing a large number of nonlinear variation characteristics, and there is redundancy and collinearity among health characteristics. Furthermore, the distribution of feature responses at different cycle stages is somewhat uneven, increasing the difficulty of feature selection. Against this backdrop, this invention chooses the random forest algorithm for feature importance evaluation primarily based on its advantages in handling nonlinear relationships, feature noise, and redundant information.

[0137] Specifically, the processing steps for constructing the optimal feature subset of the battery pack include:

[0138] S401: Calculate the aging characteristics and inconsistency characteristics of all battery packs to form an initial feature set, and associate each feature in the initial feature set with the actual SOH value of the corresponding battery pack as a sample to construct the original sample set;

[0139] S402: The bootstrap method is used to randomly select a predetermined number (two-thirds) of the samples from the original sample set as the training set to build a decision tree, and the remaining (one-third) of the samples are used as the out-of-bag data of the decision tree.

[0140] S403: The SOH prediction value of the battery pack is obtained by performing SOH prediction on the out-of-bag data (without any perturbation of the features) of each decision tree, and the mean square error between the battery pack SOH prediction value and the battery pack SOH actual value is calculated as the original out-of-bag error.

[0141] The formula is expressed as:

[0142] ;

[0143] In the formula: This represents the original bag-outside error; For the sample The predicted value; For the sample The actual SOH value of the battery pack; This represents the number of samples for out-of-bag data.

[0144] S404: Permutate all values ​​of a specific feature in the out-of-bag data of the decision tree to perturb the out-of-bag data (break the correlation between the specific feature and the actual SOH of the battery pack); re-predict the SOH of the perturbed out-of-bag data through the decision tree to obtain a new predicted SOH value for the battery pack, and calculate the mean square error between the new predicted SOH value and the actual SOH as the perturbation out-of-bag error;

[0145] S405: Calculate the difference between the perturbation out-of-bag error and the original out-of-bag error as the difference of this specific feature in the corresponding decision tree;

[0146] In this embodiment, the larger the difference, the more significant the model prediction performance declines after the association between the feature and the target value is broken, and therefore the more important the specific feature is.

[0147] S406: Repeat steps S402 to S405 to calculate the difference of the specific feature across all decision trees, and take the average of the differences of the specific feature across all decision trees as the importance score of the specific feature;

[0148] The public notice indicates that:

[0149] ;

[0150] In the formula: For specific features Importance score; The number of decision trees; This represents the original bag-outside error; To perturb the outside error of the bag;

[0151] S407: Repeat steps S402 to S406 to obtain the importance scores of all features, and sort the importance scores of all features in descending order from high to low.

[0152] S408: Select the previous step in an incremental manner The model is trained using 10 features as input features, and the accuracy of the model is recorded. A curve reflecting the change in the accuracy of the model with the number of input features is generated. The stepwise incrementing method is selected from the previous step. Each feature refers to selecting the first feature as input the first time, the first two features as input the second time, the first three features as input the third time, and so on.

[0153] S409: Identify the performance inflection point of the curve and select the number of features corresponding to that inflection point. The top-ranked candidates with the highest importance scores Each feature is used as the optimal feature subset of the battery pack. The optimal feature subset is labeled with an index for the optimal feature. In practical applications, the optimal feature is directly selected from the calculated aging features and inconsistency features based on the index of the optimal feature to form the optimal feature subset.

[0154] In this embodiment, after the performance inflection point of the curve, further increasing the number of features has very limited effect on improving the accuracy of the health status estimation model, and may even decrease it due to the introduction of noise.

[0155] In a battery pack, the characteristics of individual cells, such as capacity, internal resistance, and open-circuit voltage, form the basis of the overall performance of the battery pack. These characteristics of individual cells are accumulated or integrated within the battery pack through series and parallel configurations to form the overall characteristics of the battery pack. For example, the sum of the capacities of parallel-connected cells equals the total capacity of the battery pack, and the sum of the voltages of series-connected cells equals the total voltage of the battery pack. Therefore, the characteristics HI1 and HI2 have an additive relationship with the voltage-related characteristics of the battery pack, which can be used to construct the aging characteristics of the battery pack.

[0156] Specifically, the formula for calculating the aging characteristics of a battery pack is as follows:

[0157] ;

[0158] In the formula: For the battery pack A characteristic of aging, =1 or 2; Indicates the battery pack hierarchy. Indicates the type of health characteristic; To indicate the first The first single cell One health characteristic; Indicates the number of individual cells in the battery pack; where the battery pack has A characteristic of aging.

[0159] During the aging process, the lifespan of a battery pack is affected by internal inconsistencies. To establish a more accurate SOH estimation model for the battery pack, this invention further extracts the inconsistency characteristics of the battery pack. The difference between the health characteristics of each individual cell and its average value is defined as the inconsistency characteristic of the battery pack. The formula for calculating the inconsistency characteristic of the battery pack is as follows:

[0160] ;

[0161] In the formula: Indicates the first The first single cell The health characteristics and all individual battery cells The difference between the average values ​​of the first health characteristic, i.e., the first health characteristic of the battery pack. One inconsistent feature; Indicates the number of individual cells in the battery pack; where the battery pack has Aging characteristics. Inconsistency characteristics measure the deviation of the health status of each individual cell from the overall health status of the battery pack. The greater the deviation, the greater the difference between the individual cell's health status and the average health status of the battery pack, thus reflecting the inconsistency within the battery pack.

[0162] V. SVM-Adaboost Ensemble Model

[0163] In the specific implementation process, the SOH prediction model is the SVM-Adaboost ensemble model. SVM is suitable for processing battery health feature data with multiple features, nonlinearity, and small sample size. It is combined with Adaboost ensemble learning to build the SVM-Adaboost ensemble model to improve adaptability and performance on different datasets.

[0164] Adaboost is an ensemble learning algorithm that builds a strong classifier by combining multiple weak classifiers. It works by iteratively training the weak classifiers and adjusting the sample weights based on the performance of each classifier. Finally, the results from multiple weak classifiers are weighted and combined to improve overall classification performance.

[0165] SVM is a powerful classifier, particularly suitable for handling nonlinear problems. By mapping data to a high-dimensional space, SVM constructs a linear classification function in this high-dimensional feature space, effectively handling datasets with a large number of features. This ensures good generalization ability while solving the "curse of dimensionality" problem. SVM is well-suited for handling the multi-feature, nonlinear, and small-sample battery health feature data required in this invention. Combined with Adaboost ensemble learning, it effectively improves the model's adaptability and performance on different datasets. The core strategy of Adaboost ensemble SVM is to first use SVM as a base weak learner to build an initial prediction model. Subsequently, Adaboost integrates multiple SVM models through multiple iterations. In each iteration, Adaboost adjusts the sample weights based on the model's historical prediction performance, making the model focus more on samples with larger errors. Through this strategy, the model can be gradually optimized, thereby improving the overall regression analysis performance.

[0166] Specifically, the steps for training the SVM-Adaboost ensemble model include:

[0167] S501: Acquisition Group training samples Each training sample set includes the optimal feature subset of the battery pack and the corresponding true SOH value of the battery pack;

[0168] S502: Random selection The training dataset is composed of groups of training samples, and the sample weights of each group of training samples are set to the same initial value;

[0169] The formula is expressed as:

[0170] ;

[0171] In the formula: For the first The sample weights of each training sample;

[0172] S503: The Adaboost model's iterative loop is entered based on the current round's sample weight distribution, the optimal feature subset of the battery pack, and the battery pack's true SOH value (assuming a total of...). wheel):

[0173] S5031: In the In each iteration, a weak support vector machine learner is trained using the current sample weight distribution;

[0174] In this embodiment, the support vector machine is chosen as the ideal base model because it is good at handling nonlinear and small sample data.

[0175] S5032: Use a support vector machine weak learner to predict the training dataset and obtain the battery pack SOH prediction sequence. And calculate the weighted prediction error of the weak learner in the support vector machine. The weighted prediction error is the sum of the weights of all incorrectly predicted samples.

[0176] The formula is expressed as:

[0177] ;

[0178] In the formula: For the first In the first iteration The sample weights of each training sample;

[0179] S5033: Prediction error based on the weak learner of the support vector machine Calculate its model weights in the final strong learner. ;

[0180] In this embodiment, the weaker the prediction error, the higher its weight. The larger the entity, the greater its say in the final decision-making process.

[0181] The formula is expressed as:

[0182] ;

[0183] S5034: Update the sample weights of each training sample in the training dataset to obtain the next iteration. The sample weight distribution is determined, and the process returns to step S5031;

[0184] In this embodiment, the weight of samples that are mispredicted by the current weak learner will be increased, so that the next round of weak learners will pay more "attention" to these difficult-to-classify samples.

[0185] The formula is expressed as:

[0186] ;

[0187] In the formula: For the first In the first iteration The sample weights of each training sample; For the first In the first iteration The sample weights of each training sample; This is a normalization factor used to normalize the updated sample weights, ensuring that the sum of the sample weights is 1. For the support vector machine weak learner in the th Model weights in each iteration; For the first training samples The actual SOH value of the battery pack; For the first Round of iteration training samples The battery pack SOH prediction sequence is mainly used for training error and sample weight update;

[0188] S5035: Repeat steps S5031 to S5034 to obtain A trained support vector machine weak learner, and the model weights corresponding to each support vector machine weak learner;

[0189] S504: Based on The model weights of the support vector machine weak learners are paired. Linear fusion of multiple weak support vector machine learners yields a linear combination of multiple weak classifiers. ;

[0190] The formula is expressed as:

[0191] ;

[0192] In the formula: For the first A support vector machine weak learner is used to learn any input feature vector. The predicted values ​​are mainly used to define the final ensemble model;

[0193] S505: Based on linear combination of multiple weak classifiers Constructing a strong prediction function As a pre-trained SVM-Adaboost model;

[0194] The formula is expressed as:

[0195] ;

[0196] In the formula: This is a sign function. When the input is greater than 0, the output is 1; when the input is less than 0, the output is -1; and when the input is equal to 0, the output is 0.

[0197] To verify the estimation accuracy and stability of the proposed algorithm under different levels of inconsistency and monitoring quantities, this invention utilizes four sets of battery pack data obtained from the experimental scheme for verification. The performance of the SVM-Adaboost ensemble model is evaluated using mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE).

[0198] The calculation formula is:

[0199] ;

[0200] ;

[0201] ;

[0202] In the formula: , , These are mean absolute error, root mean square error, and mean absolute percentage error, respectively. This is the predicted SOH value for the battery pack; This represents the actual SOH value of the battery pack. This represents the number of training samples.

[0203] To verify the estimation accuracy of the established battery pack aging model under different levels of inconsistency, the first 60% of the data from the four battery packs was used as the training set and the last 40% as the test set. The optimal feature subset of the battery packs was used as the input to the SVM-Adaboost model to obtain the corresponding SOH estimate.

[0204] To verify the accuracy of the SVM-Adaboost model proposed in this invention, it was compared with several commonly used algorithms, and the particle swarm optimization (PSO) algorithm was selected to uniformly optimize and compare the algorithm parameters. The selected algorithms included PSO-SVM, PSO-GPR, and PSO-BiLSTM, and Tree-Adaboost was also introduced as an additional comparison algorithm.

[0205] Figure 2 The image shows the SOH estimation results for the battery pack. The black curve represents the actual SOH value, while the red, blue, green, purple, and brown curves represent the estimation results from the SVM-Adaboost (the model of this invention), PSO-SVM, PSO-GPR, Tree-Adaboost, and PSO-BiLSTM algorithms, respectively. Figure 2 Analysis shows that SVM-Adaboost has the best SOH estimation effect in this invention, and can effectively capture the SOH value degradation trend of the battery pack.

[0206] This invention achieves accurate estimation of the state of health (SOH) of lithium-ion battery packs with different levels of consistency by combining the extraction of features from monitorable cells, the estimation of features from unmonitored cells, and the construction of aging and inconsistency features within the group under limited monitoring conditions. It is applicable to the health status assessment of large battery packs in scenarios where data monitoring is incomplete.

[0207] As can be seen from the above technical solutions, the present invention has the following advantages:

[0208] This invention enables the estimation of the health characteristics of unmonitored cells using the health characteristics of monitorable cells without requiring comprehensive monitoring of all individual cells. Furthermore, it constructs a feature system that simultaneously characterizes the aging degree of the battery pack (i.e., intra-pack aging characteristics) and the level of inconsistency within the pack (i.e., inconsistency characteristics). By combining feature importance ranking and filtering using a random forest algorithm, the optimal features for the battery pack are selected, achieving accurate estimation of the state of health (SOH) of lithium-ion battery packs with varying degrees of consistency. This significantly improves the accuracy and robustness of SOH estimation and is suitable for assessing the health status of large battery packs in scenarios with incomplete data monitoring. Moreover, compared to existing methods that treat the battery pack as a "large battery" or rely on the deployment of all sensors, this invention balances engineering feasibility and techno-economic efficiency. It not only effectively reduces sensor configuration costs and monitoring complexity but also provides a reference for optimizing the number and placement of sensors in large battery packs.

[0209] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit the technical solutions. Those skilled in the art should understand that any modifications or equivalent substitutions to the technical solutions of the present invention without departing from the spirit and scope of the present invention should be covered within the scope of the claims of the present invention.

Claims

1. A method for estimating the state of health of lithium batteries considering inconsistencies under limited monitoring, characterized in that, include: S1: Construct several battery packs consisting of multiple individual cells connected in series and having different degrees of inconsistency; S2: Conduct battery aging tests on each battery pack to obtain the operating data of each battery pack and the corresponding actual SOH value of the battery pack; the operating data of the battery pack includes the voltage data of the monitored individual cells; S3: Perform correlation analysis on the voltage data of the monitored individual cells in each battery pack to obtain the health characteristics of the monitored individual cells, and estimate the health characteristics of the unmonitored individual cells in each battery pack based on the health characteristics of the monitored individual cells to obtain the complete health characteristic set of each battery pack. Step S3 specifically includes the following processing steps: S301: Extract candidate health features reflecting the aging state of the battery based on the voltage data of each monitored individual cell in the battery pack. S302: Calculate the Pearson correlation coefficient for the same candidate health characteristics among different monitored individual cells; S303: Select candidate health features as the health features of the monitored single cell based on the Pearson correlation coefficient; S304: Input the health features of the monitored individual cells into the trained individual cell health feature estimation model, and output the health features of the unmonitored individual cells. S305: Combine the health characteristics of the monitored individual cells and the estimated health characteristics of the unmonitored individual cells in order of cell number to obtain the complete health characteristic set of the battery pack; S4: Calculate aging characteristics and inconsistency characteristics based on the complete health feature set of each battery pack, and sort and filter the features by importance to generate the optimal feature subset of the battery pack; The index of the optimal feature is determined in the optimal feature subset of the battery pack; S5: Use the corresponding optimal feature subset of the battery pack and the true SOH value of the battery pack as a set of training samples; use the optimal feature subset of the battery pack corresponding to each training sample as the model input, and use the corresponding true SOH value of the battery pack as the supervision label to train the battery pack SOH regression prediction model. S6: Obtain the operating data of the battery pack under test, calculate the health characteristics of the monitored individual cells and estimate the health characteristics of the unmonitored individual cells to obtain a complete health feature set; calculate aging characteristics and inconsistency characteristics based on the complete health feature set of the battery pack under test, and select the optimal feature based on the index of the optimal feature to generate the optimal feature subset; input the optimal feature subset of the battery pack under test into the trained SOH prediction model, and output the SOH prediction value of the battery pack under test.

2. The lithium battery health status estimation method considering inconsistencies under limited monitoring as described in claim 1, characterized in that: In step S1, there are preset capacity differences between individual cells in each battery pack, causing each battery pack to exhibit different degrees of inconsistency.

3. The lithium battery health status estimation method considering inconsistencies under limited monitoring as described in claim 1, characterized in that: In step S2, the formula for calculating the true SOH value of the battery pack is: ； In the formula: This represents the actual SOH value of the battery pack. This represents the maximum discharge capacity of the battery pack. For the first in the battery pack The current discharge capacity of each individual cell; This refers to the rated capacity of the battery pack.

4. The lithium battery health status estimation method considering inconsistencies under limited monitoring as described in claim 1, characterized in that: In step S301, the candidate health characteristics reflecting the battery aging state include any one or more of the following: (1) Open-circuit voltage after constant current discharge; (2) The ratio of the average charging voltage to the charging capacity; (3) Area of ​​the voltage curve during constant current charging; (4) Area of ​​the voltage curve during constant voltage charging; (5) Charging cut-off voltage; (6) Charging start voltage.

5. The lithium battery health status estimation method considering inconsistencies under limited monitoring as described in claim 1, characterized in that: Step S4 specifically includes the following processing steps: S401: Calculate the aging characteristics and inconsistency characteristics of all battery packs to form an initial feature set, and associate each feature in the initial feature set with the actual SOH value of the corresponding battery pack as a sample to construct the original sample set; S402: A predetermined number of samples are randomly selected from the original sample set using the bootstrap sampling method to construct a decision tree, and the remaining samples are used as the out-of-bag data of the decision tree. S403: The SOH prediction value of the battery pack is obtained by using the decision tree to predict the SOH of the data outside the bag, and the mean square error between the predicted SOH value and the actual SOH value of the battery pack is calculated as the original outside bag error. S404: Randomly permuting all values ​​of a specific feature in the out-of-bag data of the decision tree to achieve out-of-bag data perturbation; using the decision tree to re-predict the SOH of the perturbed out-of-bag data to obtain a new SOH prediction value for the battery pack, and calculating the mean square error between the new SOH prediction value and the actual SOH value of the battery pack as the perturbation out-of-bag error; S405: Calculate the difference between the perturbation out-of-bag error and the original out-of-bag error as the difference of this specific feature in the corresponding decision tree; S406: Repeat steps S402 to S405 to calculate the difference of the specific feature across all decision trees, and take the average of the differences of the specific feature across all decision trees as the importance score of the specific feature; S407: Repeat steps S402 to S406 to obtain the importance scores of all features, and sort the importance scores of all features in descending order from high to low. S408: Select the previous one in a step-by-step incremental manner The model is trained using 10 features as input features, and the accuracy of the model is recorded to generate a curve reflecting the change in the accuracy of the model with the number of input features. S409: Identify the performance inflection point of the curve and select the number of features corresponding to the performance inflection point. ; The top with the highest importance score These features are selected as the optimal feature subset for the battery pack.

6. The lithium battery health status estimation method considering inconsistencies under limited monitoring as described in claim 5, characterized in that: In step S401, the formula for calculating the aging characteristics of the battery pack is: ； In the formula: For the battery pack A characteristic of aging, Indicates the battery pack hierarchy. Indicates the type of health characteristic; To indicate the first The first single cell One health characteristic; Indicates the number of individual cells in the battery pack; where the battery pack has A characteristic of aging.

7. The lithium battery health status estimation method considering inconsistencies under limited monitoring as described in claim 5, characterized in that: In step S401, the formula for calculating the inconsistency characteristics of the battery pack is: ； In the formula: Indicates the first The first single cell The health characteristics and all individual battery cells The difference between the average values ​​of the first health characteristic, i.e., the first health characteristic of the battery pack. One inconsistent feature; Indicates the number of individual cells in the battery pack; where the battery pack has A characteristic of aging.

8. The lithium battery health status estimation method considering inconsistencies under limited monitoring as described in claim 1, characterized in that: In step S5, the SOH prediction model is the SVM-Adaboost ensemble model; The steps involved in training the SVM-Adaboost ensemble model include: S501: Acquisition Each training sample set includes the optimal feature subset of the battery pack and the corresponding true SOH value of the battery pack. S502: Random selection The training dataset is composed of groups of training samples, and the sample weights of each group of training samples are set to the same initial value; S503: The Adaboost model's iterative loop is entered based on the current round's sample weight distribution, the optimal feature subset of the battery pack, and the battery pack's true SOH value. S5031: In the In each iteration, a weak support vector machine learner is trained using the current sample weight distribution; S5032: Use a support vector machine weak learner to predict the training dataset and obtain the battery pack SOH prediction sequence. And calculate the weighted prediction error of the weak learner in the support vector machine. ; The formula is expressed as: ； In the formula: For the first In the first iteration The sample weights of each training sample; S5033: Weighted prediction error based on the weak learner of the support vector machine. Calculate its model weights in the final strong learner. ; The formula is expressed as: ； S5034: Update the sample weights of each training sample in the training dataset to obtain the next iteration. The sample weight distribution is determined, and the process returns to step S5031; The formula is expressed as: ； In the formula: For the first In the first iteration The sample weights of each training sample; For the first In the first iteration The sample weights of each training sample; This is a normalization factor used to normalize the updated sample weights. For the support vector machine weak learner in the th Model weights in each iteration; For the first training samples The actual SOH value of the battery pack; For the first Round of iteration training samples The sequence of predicted SOH values ​​for the battery pack; S5035: Repeat steps S5031 to S5034 to obtain A trained support vector machine weak learner, and the model weights corresponding to each support vector machine weak learner; S504: Based on The model weights of the support vector machine weak learners are paired. Linear fusion of multiple weak support vector machine learners yields a linear combination of multiple weak classifiers. ; The formula is expressed as: ； In the formula: For the first A support vector machine weak learner is used to learn any input feature vector. The predicted value; S505: Based on linear combination of multiple weak classifiers Constructing a strong prediction function As a pre-trained SVM-Adaboost model; The formula is expressed as: ； In the formula: This is a sign function. When the input is greater than 0, the output is 1; when the input is less than 0, the output is -1; and when the input is equal to 0, the output is 0.

9. The lithium battery health status estimation method considering inconsistencies under limited monitoring as described in claim 8, characterized in that: In step S5, the performance of the SVM-Adaboost ensemble model is evaluated using mean absolute error, root mean square error, and mean absolute percentage error. The calculation formula is: ； ； ； In the formula: , , These are mean absolute error, root mean square error, and mean absolute percentage error, respectively. This is the predicted SOH value for the battery pack; This represents the actual SOH value of the battery pack. This represents the number of training samples.

Images