Lithium ion battery SOH prediction method

Abstract

The invention discloses a lithium ion battery SOH (state of health) prediction method, which comprises the following steps of: acquiring a lithium ion battery cyclic charge-discharge data set, and training an SOH prediction model, the lithium ion battery cyclic charge-discharge data set comprising charge-discharge data of a fast charge stage and a constant voltage stage under multiple charge-discharge cycles; according to the lithium ion battery cycle charging and discharging data set, feature sets used for representing battery health factors are constructed for different charging stages; based on the feature set of each charging stage, high-correlation features are screened out; inputting a feature set corresponding to the high-correlation features into a pre-constructed regression model for training; predicting the SOH of the lithium ion battery according to the trained SOH prediction model; according to the method, the lithium ion battery cycle charging and discharging data set is utilized, the charging process of the lithium ion battery is analyzed, the health factors reflecting the SOH are extracted, the prediction model is established, and accurate prediction of the SOH of the lithium ion battery can be achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of battery life prediction, and more particularly to a method for predicting SOH of a lithium-ion battery. Background Art

[0002] At present, lithium-ion batteries have been widely used in electric vehicles and energy storage. Lithium-ion batteries have ideal energy density and cycle life and are currently the best choice for these applications. As the development of lithium-ion batteries tends towards higher energy density and more complex battery dynamics, battery efficiency and safety will become a key issue, which mainly depends on the accurate estimation of the battery state of health (SOH). In more detail, SOH indicates the degree of battery aging, which is defined as the ratio of actual capacity to rated capacity. This parameter gradually decreases with the cycle of the battery, and when it drops to 80%, the battery life is over.

[0003] Generally speaking, lithium battery SOH estimation methods can be roughly divided into two categories. The first category is model-based methods, which mainly reorganize equivalent circuit models and electrochemical models. Among them, the equivalent circuit model method has high computational efficiency and is suitable for online battery state prediction, but how to obtain higher accuracy remains a challenge. The electrochemical model is superior to the equivalent circuit model method in terms of intrinsic physical characterization, but its disadvantage is that it is limited to specific battery systems. The second category is data-driven methods, which regard SOH estimation as a typical regression problem and use machine learning methods to map the nonlinear relationship between input data and battery SOH. The advantage of data-driven is that it does not require analysis of internal mechanisms and can achieve accurate predictions in the presence of large amounts of data.

[0004] The current SOH prediction methods have some defects, such as the inability to make accurate predictions when faced with complex charging conditions.

[0005] Therefore, how to achieve accurate prediction of SOH of lithium-ion batteries is an urgent problem to be solved by technicians in this field. Summary of the invention

[0006] In view of this, the present invention provides a lithium-ion battery SOH prediction method, which can achieve accurate prediction of the lithium-ion battery SOH by using the lithium-ion battery cycle charge and discharge data set.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] A method for predicting SOH of a lithium-ion battery comprises the following steps:

[0009] Obtaining a lithium-ion battery cyclic charge and discharge data set and training a life prediction model: wherein the lithium-ion battery cyclic charge and discharge data set includes charge and discharge data in a fast charge phase and a constant voltage phase under multiple charge and discharge cycles;

[0010] According to the lithium-ion battery cycle charge and discharge data set, constructing feature sets for characterizing battery health factors for different charging stages;

[0011] Inputting the feature set into a pre-built regression model for training;

[0012] Predict the SOH of lithium-ion batteries based on the trained lifetime prediction model.

[0013] Preferably, the charge and discharge data in the fast charge stage includes voltage change data of each cycle under constant capacity.

[0014] Preferably, the feature set constructed for the fast charging stage includes:

[0015] HI 1,m =V n -V1,m=linspace(1,100,100)

[0016]

[0017]

[0018]

[0019] Among them, HI 1,m , HI 2,m , HI 3,m , HI 4,m They represent the different health factor characterization characteristics of the fast charging stage; n represents the nth cycle; m represents the mth capacity node; Vn represents the voltage of the nth cycle; V1 represents the voltage of the 1st cycle; linspace represents linear equal division; Represents the mean of all pressure differences of the sample.

[0020] Preferably, the charge and discharge data in the constant voltage stage includes the capacity corresponding to each cycle under constant voltage.

[0021] Preferably, the feature set constructed for the constant pressure stage includes:

[0022] HI 5,m =Q n ,m=linspace(1,50,50)

[0023]

[0024]

[0025]

[0026] Among them, HI 5,m , HI 6,m , HI 7,m , HI 8,m They represent the different health factor characteristics of the CV charging stage; Q n represents the voltage of the nth cycle; Represents the mean of all the sample sizes.

[0027] Preferably, the step further includes: performing a correlation analysis on the feature set, calculating the correlation between each feature in the feature set and the SOH through the correlation analysis, and screening out features whose correlations meet conditions to be used as inputs of the regression model.

[0028] Preferably, the step further includes: analyzing the model error according to the predicted value of the model and the corresponding true value, and the error calculation formula of the model is:

[0029]

[0030] Where n represents the number of cycles of the battery; y real represents the true value; y predict represents the predicted value; ΔS represents the integral difference between the two curves.

[0031] A method for constructing a training data set for lithium battery SOH prediction includes the following steps:

[0032] Obtain the capacity and voltage data corresponding to multiple cycles of charge and discharge of lithium-ion batteries;

[0033] Dividing the capacity and voltage data into fast charging stage data and constant voltage stage data;

[0034] According to the fast charging stage data and constant voltage stage data under multiple cycles, the feature sets of the corresponding charging stages are constructed respectively;

[0035] Based on the feature sets of each charging stage, highly correlated features are screened out, and the feature sets corresponding to the highly correlated features are used as training data sets.

[0036] A lithium-ion battery SOH prediction system comprises a computer system, wherein the computer system is used to execute the lithium-ion battery SOH prediction method.

[0037] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned lithium-ion battery SOH prediction method is implemented.

[0038] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method for predicting SOH of a lithium-ion battery, which can accurately predict the SOH of a lithium-ion battery by using a lithium-ion battery cycle charge and discharge data set. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0040] Figure 1 A schematic diagram of a lithium-ion battery SOH prediction method provided by an embodiment of the present invention;

[0041] Figure 2 A charge and discharge cycle curve diagram of a lithium battery in an embodiment of the present invention;

[0042] Figure 3 Schematic diagram of battery characteristics during the charging process of a lithium-ion battery;

[0043] Figure 4 It is a heat map of the correlation between the characteristics of each health factor and SOH in the embodiment of the present invention;

[0044] Figure 5 Schematic diagram of the prediction effect of the GPR life prediction model in an embodiment of the present invention;

[0045] Figure 6 Schematic diagram of prediction error analysis results of GPR life prediction in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] Example 1

[0048] like Figure 1 The embodiment of the present invention discloses a method for predicting SOH of a lithium-ion battery, comprising the following steps:

[0049] S1: Obtain a lithium-ion battery cycle charge and discharge dataset and train a SOH prediction model.

[0050] S2: Input the feature set into the pre-built regression model for training.

[0051] The inventive concept of the present invention is to construct a charging and discharging data set training model under complex working conditions so that the trained model can have better recognition ability, thereby improving the accuracy of SOH prediction.

[0052] In this embodiment, the present invention divides the entire charging process into different charging stages to obtain and construct training data sets. In the charging process of lithium-ion batteries, there are three stages according to the charging mode: a primary fast charging stage, a secondary fast charging stage, and a constant voltage (CV) charging stage.

[0053] like Figure 2 There is no fixed boundary between the first and second fast charging stages. The whole process lasts from the beginning until the battery SOC reaches 80%. During this period, constant current (CC) charging is performed, and then the battery is charged with CV at a charging rate of 1C until it is fully charged. During the entire discharge process, all batteries are discharged at a constant current of 4C charging rate until the power is fully released.

[0054] The CC-CV (Constant Current-Constant Voltage) charging mode is a charging strategy commonly used for lithium-ion batteries, which aims to improve charging efficiency and protect battery health. In the CC stage, the battery is charged at a constant current until the battery voltage reaches a preset upper limit. Subsequently, in the CV stage, the battery voltage is maintained at this upper limit, and the current gradually decreases as the battery charging state improves until charging is completed.

[0055] CC stage: In the constant current charging stage, the charging system applies a constant current to the battery to achieve a faster charging speed. In this stage, the battery voltage continues to increase until it reaches the predicted maximum voltage. Due to the existence of polarization, the battery voltage measured during the charging process may be slightly higher than the actual battery voltage.

[0056] CV stage: In the constant voltage charging stage, the charging system keeps the charging voltage constant, the current decreases exponentially, and the polarization voltage gradually decreases. In this stage, the measured battery voltage is closer to the actual voltage of the battery. When the charging current decreases to a certain value or the battery SOC increases to a certain value, it can be considered that the battery is fully charged and the charging process is over.

[0057] The data set used in the present invention consists of multiple lithium-ion batteries that are cycled to the end of their life under multi-stage fast charging conditions, and their life cycles range from 200 to 2500 times. The first-level fast charging stage and the second-level fast charging stage are unified as the fast charging stage. Therefore, each cycle will go through the fast charging stage and the CV charging stage, thereby obtaining the charge and discharge data of the fast charging stage and the CV stage under multiple charge and discharge cycles.

[0058] In addition, in order to make the trained model have stronger generalization ability and higher accuracy, when selecting the lithium-ion battery to generate the data set, the range of lithium-ion battery life can be widened as much as possible, such as 500-2000 times.

[0059] In order to further implement the above technical solution, after obtaining the charge and discharge data set under multiple cycles, the battery health factor of the fast charging stage and the battery health factor of the CV charging stage are extracted by extracting the voltage and capacity changes as the feature sets of the corresponding stages.

[0060] Among them, the battery health factor (HI) is a quantitative parameter used to indicate the health of the battery relative to its new state. It is usually calculated by analyzing multiple performance parameters of the battery (such as capacity, internal resistance, voltage, etc.) and taking into account the changing trends of these parameters over time. The higher the HI value, the better the health of the battery, and vice versa.

[0061] like Figure 3 , which are the VQ curve and QV curve of the lithium-ion battery charging process. As can be seen from the figure, with the increase of the number of cycles, the VQ curve continues to move upward.

[0062] Figure 3 Figure (a) is the VQ curve of the fast charging stage, and the battery health factor of the fast charging stage is extracted based on the VQ curve. It can be found from Figure (a) that the fast charging stage has two charging modes. When the charging mode is switched, the voltage will drop slightly. Therefore, when extracting the battery health factor for the fast charging stage, this embodiment uses the voltage change under constant capacity as a characteristic that responds to the battery capacity decay, takes Q as the independent variable, selects multiple capacity nodes during the fast charging stage, and calculates the voltage difference generated by the number of cycles at each capacity node to confirm the battery health factor. The battery health factor can be obtained by one or more of the following extraction methods.

[0063] Method 1: HI 1,m =V n -V1,m=linspace(1,100,100).

[0064] Method 2:

[0065] Method 3:

[0066] Method 4:

[0067] Explain the above four methods: Among them, HI 1,m , HI 2,m , HI 3,m , HI 4,m They represent the health factor characterization features extracted by different methods in the fast charging stage. n represents the nth cycle; m represents the mth capacity node; Vn represents the voltage of the nth cycle; V1 represents the voltage of the 1st cycle; linspace represents linear equal division; Represents the mean of all pressure differences of the sample.

[0068] Method 1 means dividing the fast charging process into 100 segments, that is, in the fast charging process of each battery, HI is extracted according to the pressure difference corresponding to the 100 capacity nodes; similarly, method 2 means dividing the fast charging process into 10 segments and extracting HI according to the pressure difference; method 3 means dividing the fast charging process into 10 segments and extracting HI according to the mean of the cumulative pressure difference; method 4 means dividing the fast charging process into 10 segments and extracting HI according to the standard deviation of the cumulative pressure difference.

[0069] Figure 3 Figure (b) is the QV curve of the CV charging stage, and the battery health factor of the CV stage is extracted based on the QV curve. As can be seen from Figure (b), in the CV stage, as the battery life decays, the capacity continues to decrease. When Q is used as the independent variable, the battery health factor cannot be extracted as the capacity decreases. Therefore, when extracting the battery health factor in the CV stage, this embodiment uses the voltage V as the independent variable, selects multiple voltage nodes during the CV charging process, and calculates the corresponding capacity of different cycle times at each voltage node to confirm the battery health factor. The battery health factor can be obtained by one or more of the following extraction methods.

[0070] Method 1: HI 5,m =Q n ,m=linspace(1,50,50).

[0071] Method 2:

[0072] Method 3:

[0073] Method 4:

[0074] Explain the above four methods: Among them, HI 5,m , HI 6,m , HI 7,m , HI8,m They represent the health factor characterization features extracted by different methods in the fast charging stage. n represents the nth cycle; m represents the mth node, which is the voltage node for the CV charging stage; Q n represents the capacity of the nth cycle; Represents the mean of all the sample sizes.

[0075] Furthermore, the characterization features of health factors at each stage are taken as feature sets, and correlation analysis is performed based on the feature sets to obtain more meaningful features to improve the interpretability of the model.

[0076] Figure 4 The heat map of the correlation between the characteristics of each health factor and SOH, where (a)-(d) are HI in the fast charging stage 1,m , HI 2,m , HI 3,m , HI 4,m Correlation analysis results with SOH; (e)-(j) are HI in the constant pressure stage 5,m , HI 6,m , HI 7,m , HI 8,m The results of the correlation analysis with SOH. The vertical axis in the figure represents the number of the sample batteries, and the cycle life is arranged from low to high. The horizontal axis in the figure represents the nodes selected by the corresponding extraction method. Blue represents a higher negative correlation, and yellow represents a higher positive correlation.

[0077] The analysis method of correlation can be Spearman analysis, and the monotonic relationship between two variables can be evaluated by the Spearman correlation coefficient. The expression of the Spearman correlation coefficient is:

[0078]

[0079] In the formula, α and β represent two variables. and represents their average value.

[0080] In addition, other correlation analysis methods can also be used, such as Pearson analysis, Kendall analysis, etc. An ideal HI should ensure that the SOH of all batteries at the same node is highly correlated. Figure 4 (e) and (f) can meet this condition (depending on the extraction method, they can also be considered as the same HI). Figure 4 (c) and (d) are also roughly satisfied, but upon closer inspection, Figure 4The correlation of some batteries in (c) and (d) cannot be maintained at the highest level. In this experiment, for each battery, since HI5 and SOH can maintain a high correlation, it is finally used as the input of the subsequent model to predict SOH. If other types of batteries are replaced or the experimental conditions are changed, specific analysis is required to select the corresponding high correlation features.

[0081] In this experiment, complex working conditions are mainly reflected in the CC (constant current) charging stage, in which high-rate charging or low-rate charging can be performed according to demand. Since the selected health factor is set in the CV (constant voltage) stage, the charging process (low current charging) in the CV stage is stable and is not affected by the CC stage, so the health factor extracted in the CV stage can identify complex working conditions.

[0082] In order to further implement the above technical solution, the life prediction module in this embodiment is established using the Gaussian regression (GPR) algorithm. GPR is usually used to approximate the target output f(x), where x is a d-dimensional n-input vector, and the formula is as follows:

[0083] f(x)~GP(m(x),k f (x,x'))

[0084] m(x)=E[f(x)]

[0085] k f (x,x')=E[(f(x)-m(x))(f(x')-m)(x') T ]

[0086] Among them, m(x) represents the mean function; k f (x, x') represents the covariance function, i.e., the kernel function, and x' and x represent different input vectors.

[0087] In this embodiment, the training features confirmed after correlation analysis and the corresponding SOH labels are input into the GPR life prediction model for training. The GPR model constructs a covariance function based on the training features and confirms the prior distribution in combination with the SOH labels, and optimizes the hyperparameters through the likelihood function to complete the model training. The specific steps are as follows:

[0088] First, confirm that the kernel function is:

[0089]

[0090] Among them, σ m represents the characteristic length scale of the m-th predictor; represents the signal variance; α Represents a positive value to scale the blending parameter.

[0091] For most practical situations, the SOH label can be expressed as an implicit function as follows:

[0092] y=f(x)+ε

[0093] Where y represents the observation vector; ε Represents noise.

[0094] The prior distribution of the observed values ​​is expressed as:

[0095] y=N(0,Cov(x,x)+δ 2 I n )

[0096] Among them, I n represents the n-dimensional identity matrix; δ 2 I n represents the noise covariance matrix.

[0097] Finally, the hyperparameter set [σ f ,σ m ,α].

[0098] Furthermore, after completing the training of the GPR model through the above steps, the posterior distribution is used for prediction. In the test set, the predicted value y * Also obeys Gaussian distribution, at the test point x * At the point where the observed value y and the predicted value y * The joint prior distribution of is expressed as:

[0099]

[0100] Consider a joint Gaussian prior distribution on y, y * The posterior distribution of is expressed as:

[0101] p(y * |x,y,x * )=N(μ,Σ)

[0102] μ=Cov(x * ,x)[Cov(x,x)+δ 2 I n ] -1 y

[0103] Σ=Cov(x * ,x * )-Cov(x * ,x)[Cov(x,x)+δ 2 I n ] -1 Cov(x,x* )

[0104] In the formula, μ represents the estimated value of the test set; Σ represents the covariance prediction value. Σ is used to reflect the uncertainty of the GPR model.

[0105] To further illustrate the technical effects of the present invention, please refer to Figure 5 , Figure 5 The actual prediction effect of the GPR life prediction model is shown. Figures (a), (b) and (c) are the battery life prediction results for the low life segment, medium life segment and high life segment respectively. From the prediction results, it can be seen that the model can accurately predict batteries under different operating conditions (different life segments) and has high prediction accuracy and generalization ability.

[0106] Furthermore, in order to verify the rationality of the model, all batteries except the training set are verified. The present invention adopts a new error analysis method, and the calculation formula is as follows:

[0107]

[0108] Where n represents the number of cycles of the battery; y real represents the true value; y predict represents the predicted value; ΔS represents the integral difference between the two curves.

[0109] Attached Figure 6 To analyze the error results, the present invention uses a variety of evaluation indicators to analyze the prediction error of the model. Figure 6 The evaluation indicators used in Figures (a), (b), (c), and (d) are Mean Absolute Percentage Error (MAPE), Root Mean Squared Error (RMSE), Coeffcient of Determination (R 2 ) and ΔS.

[0110] from Figure 6 As can be seen in (d), for ΔS, it is generally maintained near 0, and there is no obvious trend change. The above results and analysis can prove that the model can achieve accurate prediction of battery SOH under different operating conditions and has strong generalization ability.

[0111] In summary, when faced with complex charging conditions and limited data, we can choose to broaden the selection range of the training set and use low-, medium-, and long-life batteries as training sets to build a model. This will increase the power and improve the prediction accuracy. The MAPE of the model established by the present invention can reach 0.45%.

[0112] Example 2

[0113] Based on the same inventive concept, an embodiment of the present invention discloses a method for constructing a training data set for lithium battery SOH prediction, comprising the following steps:

[0114] Obtain the capacity and voltage data corresponding to multiple cycles of charge and discharge of lithium-ion batteries;

[0115] Divide the capacity and voltage data into fast charging stage data and constant voltage stage data;

[0116] According to the fast charging stage data and constant voltage stage data under multiple cycles, the feature sets of the corresponding charging stages are constructed respectively;

[0117] Based on the feature sets of each charging stage, highly correlated features are screened out, and the feature sets corresponding to the highly correlated features are used as training data sets.

[0118] In this embodiment, multiple lithium-ion batteries are selected for cyclic charge and discharge, and after multiple cycles until the end of the life span, the change curves of battery capacity and battery voltage are recorded during each charging process. As the cycle progresses, the battery capacity and battery voltage curves also change accordingly, so features related to battery life can be extracted.

[0119] In order to further implement the above technical solution, for the fast charging stage, the voltage change characteristics under constant capacity are extracted as the battery health factor; for the CV stage, the capacity change characteristics under constant voltage are extracted as the battery health factor.

[0120] Among them, one or more of the following extraction methods are used to obtain the battery health factor in the fast charging stage:

[0121] Method 1: HI 1,m =V n -V1,m=linspace(1,100,100)

[0122] Method 2:

[0123] Method 3:

[0124] Method 4:

[0125] Among them, HI 1，m , HI 2，m , HI 3，m , HI 4，mThey represent the health factor characterization features extracted by different methods in the fast charging stage; n represents the nth cycle; m represents the mth capacity node; Vn represents the voltage of the nth cycle; V1 represents the voltage of the 1st cycle; linspace represents linear equal division; Represents the mean of all pressure differences of the sample.

[0126] One or more of the following extraction methods are used to obtain the battery health factor in the CV stage:

[0127] Method 1: HI 5,m =Q n , m=linspace(1,50,50).

[0128] Method 2:

[0129] Method 3:

[0130] Method 4:

[0131] Among them, HI 5，m , HI 6，m , HI 7，m , HI 8，m They represent the different health factor characteristics of the CV charging stage; Q n represents the capacity of the nth cycle; Represents the mean of all the sample sizes.

[0132] Example 3

[0133] Based on the same inventive concept, an embodiment of the present invention discloses a lithium-ion battery SOH prediction system, including a computer system. The computer system can perform SOH prediction of lithium ions in the manner of embodiment 1.

[0134] Example 4

[0135] Based on the same inventive concept, an embodiment of the present invention discloses a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the lithium-ion battery SOH prediction method in embodiment 1 can be implemented.

[0136] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0137] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A lithium-ion battery SOH prediction method, characterized in that: The following steps are involved: Obtaining a lithium-ion battery cyclic charge and discharge data set and training a life prediction model: wherein the lithium-ion battery cyclic charge and discharge data set includes charge and discharge data in a fast charge phase and a constant voltage phase under multiple charge and discharge cycles; According to the lithium-ion battery cycle charge and discharge data set, constructing feature sets for characterizing battery health factors for different charging stages; Based on the feature sets of each charging stage, high-relevance features are screened out; Input the feature set corresponding to the highly correlated features into the pre-built regression model for training; Predict the SOH of lithium-ion batteries based on the trained SOH prediction model.

2. A lithium-ion battery SOH prediction method according to claim 1, characterized in that: The charge and discharge data in the fast charge stage include voltage change data of each cycle under constant capacity.

3. A lithium-ion battery SOH prediction method according to claim 2, characterized in that: The feature set constructed for the fast charging stage includes: obtaining the battery health factor by using one or more extraction methods in the following ways: Method 1: HI 1,m =V n -V1,m=linspace(1,100,100) Method 2: Method 3: Method 4: Among them, HI 1,m , HI 2,m , HI 3,m , HI 4,m They represent the health factor characterization features extracted by different methods in the fast charging stage; n represents the nth cycle; m represents the mth capacity node; Vn represents the voltage of the nth cycle; V1 represents the voltage of the 1st cycle; linspace represents linear equal division; Represents the mean of all pressure differences of the sample.

4. A lithium-ion battery SOH prediction method according to claim 1 or 2, characterized in that: The charge and discharge data in the constant voltage stage include the capacity corresponding to each cycle under constant voltage.

5. A lithium-ion battery SOH prediction method according to claim 4, characterized in that: The feature set constructed for the constant voltage stage includes: obtaining the battery health factor by using one or more of the following extraction methods: Method 1: HI 5,m =Q n ,m=linspace(1,50,50) Method 2: Method 3: Method 4: Among them, HI 5,m , HI 6,m , HI 7,m , HI 8,m They represent the different health factor characteristics of the constant voltage charging stage; Q n represents the capacity of the nth cycle; Represents the mean of all the sample capacities.

6. A lithium-ion battery SOH prediction method according to claim 1, characterized in that: The step also includes: performing a correlation analysis on the feature set, calculating the correlation between each feature in the feature set and the SOH through the correlation analysis, and screening out features that meet the correlation conditions to be used as inputs of the regression model.

7. A lithium-ion battery SOH prediction method according to claim 1 or 6, characterized in that: The steps also include: analyzing the model error according to the predicted value of the model and the corresponding true value, and the error calculation formula of the model is: Where n represents the number of cycles of the battery; y real represents the true value; y predict represents the predicted value; ΔS represents the integral difference between the two curves.

8. A method for constructing a training data set for lithium battery SOH prediction, characterized in that , including the following steps, Obtain the capacity and voltage data corresponding to multiple cycles of charge and discharge of lithium-ion batteries; Dividing the capacity and voltage data into fast charging stage data and constant voltage stage data; According to the fast charging stage data and constant voltage stage data under multiple cycles, the feature sets of the corresponding charging stages are constructed respectively; Based on the feature sets of each charging stage, highly correlated features are screened out, and the feature sets corresponding to the highly correlated features are used as training data sets.

9. A lithium-ion battery SOH prediction system, characterized in that: The invention comprises a computer system, wherein the computer system is used to execute the lithium-ion battery SOH prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the lithium-ion battery SOH prediction method according to any one of claims 1 to 7 is implemented.

Images