Power battery and energy storage battery health state prediction method and system

Abstract

The invention relates to a power battery and energy storage battery health state prediction method and system, and the method comprises the following steps: S1, obtaining original EIS data through an open source data set, extracting the center position, radius and arc length of a medium-high frequency Nyquist diagram arc, and extracting the length of a low-frequency Nyquist diagram as health features, and obtaining a training set; step S2, constructing a DGPR model; inputting the training set into the DGPR for training, selecting a kernel function, training to obtain a hyper-parameter, and establishing an accurate battery health state prediction model; and S3, inputting the health features extracted in the test set into the trained health state estimation model, and performing battery health state prediction according to hyper-parameters obtained by training. According to the method, the health features can be extracted by intercepting the EIS frequency band data, the health features are input into the DGPR for training to obtain the hyper-parameters, and the health state of the battery is predicted according to the hyper-parameters.

Description

Technical Field

[0001] The present invention relates to the field of battery health prediction, in particular to a method and system for predicting the health state of power batteries and energy storage batteries. Background Art

[0002] This patent relates to the fields of power batteries and energy storage batteries, including lithium iron phosphate batteries, ternary lithium batteries, and graphene batteries; in particular, it relates to a method and system for predicting the health state of batteries based on the geometric characteristics of the SEI film process and Gaussian regression process characterized by electrochemical impedance spectroscopy. A battery is an electrochemical device with a complex electrochemical structure and nonlinearity. Its internal materials, aging degree, external temperature and other factors seriously affect the health state of the battery. On the one hand, during the production process of new energy batteries, there are differences in the health states of batteries produced in the same batch. This patent innovatively proposes a new detection method for consistency indicators, which can be graded according to the actual battery health state gradient to ensure the consistency performance of battery packs in the same gradient in high-demand applications. On the other hand, when the battery undergoes repeated charge / discharge cycles and deteriorates to an unacceptable level, the battery system will have catastrophic failures, which is not conducive to the safe use and recycling of the battery. For this reason, this patent provides a technical means for detecting the safety of the battery, which can recommend retiring the battery when its health state is 80% and using it in application scenarios with low-performance requirements, such as using retired batteries of new energy electric vehicles in electric two-wheelers, etc. At the same time, by applying the retired batteries in a hierarchical and echelon manner, the utilization value of the retired batteries can be improved. This patent can provide effective technical monitoring and screening means for the battery retirement and recycling industry. The differences in battery materials and aging degrees pose challenges to the prediction of the battery health state. Accurately evaluating the health state of the battery helps to predict the battery operation range, improve the use performance and safety; in terms of battery recycling, the level of the health state not only determines the feasibility of battery reuse, but also directly affects the economic value and environmental benefits of the battery.

[0003] In previous patents, the main prediction methods for the state of health of batteries focused on establishing battery degradation models and data-driven approaches. The model-based method first establishes a mathematical model to capture the dynamic behavior of the battery, such as SEI film growth, charge transfer process, degradation process, etc. Then, parameter identification and state observation algorithms are designed to online monitor health-related parameters, such as capacitance and internal resistance. The battery degradation characteristics are affected by multiple stress factors, including depth of discharge, temperature, load profile, and battery state of charge range, etc. These influencing factors exhibit non-linear and coupling effects on the battery degradation behavior, so it is difficult to establish an accurate degradation model to capture the battery degradation dynamics; moreover, on-site charge and discharge tests are required to calibrate the state of health of the battery, which results in long test time, low efficiency, and high cost, directly affecting the economic feasibility of second-life utilization. The data-driven method collects battery operation data (such as voltage, current, temperature, EIS) and uses machine learning models to extract features and establish the mapping relationship between the state of health of the battery and the features to achieve accurate prediction. Electrochemical impedance spectroscopy (EIS) reflects the impedance information of the battery in a wide frequency range, including ohmic resistance, SEI film resistance, charge transfer resistance, and diffusion resistance. They change with the aging of the battery, so EIS helps to describe the degree of battery degradation. The latest approach is usually to obtain battery health features through EIS and obtain the mapping relationship with the state of health of the battery through machine learning to predict the state of health of the battery. Identifying and extracting reliable health indicators (HI) has become the main bottleneck in adopting the data-driven method. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for predicting the state of health of power batteries and energy storage batteries, which can extract health features by intercepting EIS frequency band data, input the health features into DGPR for training to obtain hyperparameters, and predict the state of health of the battery according to the hyperparameters.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A method for predicting the state of health of power batteries and energy storage batteries, characterized in that: the method includes the following steps:

[0006] Step S1: Obtain the original EIS data through an open-source dataset, and extract the center position, radius, and arc length of the mid-high frequency Nyquist plot arc, as well as the length of the low-frequency Nyquist plot as health features to obtain a training set;

[0007] Step S2: Construct a DGPR model; input the training set into DGPR for training, select a kernel function, train to obtain hyperparameters, and establish an accurate battery state of health prediction model;

[0008] Step S3: Input the health features extracted from the test set into the trained health state estimation model, and predict the battery health state according to the hyperparameters obtained by training.

[0009] Further, define the health feature set X = [x1,..., x n T and the battery health state set Y = [y1,..., y n T , where the training set in step S1 is {(x i , y i ), i = 1, 2,..., n}. The input value x i = [HI1, HI2, HI3], including the weighted radius calculated by fitting the real and imaginary parts of the impedance spectrum collected at 20KHz to 45Hz in the current cycle to the center coordinate position and radius of the semi-circle, as well as the semi-circle arc length and the shape length of the diffusion process EIS Nyquist plot. The output value y i is the battery health state corresponding to the health features. Normalize the real and imaginary parts of the EIS data set using the mean and standard deviation of the data in the training set, and the EIS sequence data in the open source data set.

[0010] The acquisition method of the health features in step S1 is as follows:

[0011] Use the real and imaginary part data EIS of the EIS data in the frequency range of [20kHz - 45Hz] with a total of 60 frequencies 1-26 to calculate the center position and radius of the first semi-circle of the Nyquist plot, as well as the polar coordinates of the center. The Nyquist plot of the battery EIS is close to a semi-circle in the medium and high frequency range. Therefore, the EIS characteristics in the medium and high frequency range can be described as:

[0012] (Z′ - X) 2 +(Z″ - Y) = R 2

[0013] where EIS 1-26 represents the data corresponding to the 1st to 26th frequencies in the 60-frequency (from large to small) data of a certain sequence of EIS data in the data set. Z′ and Z″ are the real and imaginary part data of the EIS; X and Y are the center positions of the Nyquist semi-circle, and R is the radius of the semi-circle. For each data point, its geometric distance to the center is:

[0014]

[0015] The objective function for fitting the semi-circle is:

[0016]

[0017] ​​Use the gradient descent method to obtain X, Y, and R. The formula for Health Feature 1 is:

[0018]

[0019] Obtain the included angle θ between the two end points of the selected frequency band and the center of the semi - circle by calculation, so as to obtain the arc length corresponding to the semi - circle fitted by the selected frequency band. The formula for Health Feature 2 is:

[0020] HI2 = R×θ

[0021] Obtain the EIS 27-60 The EIS data corresponding to the minimum value of the imaginary part in the EIS is the starting point (S) of the battery electrochemical diffusion process. The EIS 60 is used as the end point (E). The formula for Health Feature 3 is:

[0022]

[0023] Among them, S z′ represents the real part of the EIS corresponding to the starting point, and S z″ represents the imaginary part of the starting point; EIS 27-60 represents the data corresponding to the 27th to 60th frequencies (from large to small) in the 60 - frequency EIS data of a certain sequence in the data set. The EIS 60 represents the data at the 60th frequency point of the EIS data of a certain sequence in the data set; E Z' represents the real part of the end point of the diffusion process, and E Z” represents the imaginary part of the end point of the diffusion process.

[0024] Furthermore, the step S2 is:

[0025] Given a training set τ = {(x i and output y i ), i = 1, 2, 3,..., n} composed of n pairs of input x i and output y i ), calculate the predicted distribution of the unknown observed value y * at x * . The GPR assumes that any set of function values f(x) = [f(x1), …, f(x n )] follows a multivariate Gaussian distribution, that is:

[0026]

[0027] where K is a matrix, and its elements are K ij = k(x i , x j ), and k(·,·) is the kernel function;

[0028] Input the training set into DGPR for training, and select MaternARD3 / 5 As the kernel function, the hyperparameters hyp = [lik, cov1, cov2,..., cov n are obtained through training.

[0029] lik represents the logarithmic standard deviation of the observation noise, describing the magnitude of the noise in the data.

[0030] cov1 describes the variation range of the input features, indicating how close two data points need to be considered relevant.

[0031] cov2 - n describes the magnitude of the output value of the kernel function, that is, the overall variation range of the data. Using 35C01 / 35C02 as the test set, the state of health of the battery is predicted according to the trained hyperparameters.

[0032] Furthermore, the step S3 is as follows:

[0033] GPR performs non - parametric regression using the Gaussian process, assuming is independent and identically - distributed Gaussian noise, and the output f=(f(x1),..., f(x n )) is modeled as a Gaussian random field

[0034] where K ij =k(x i , y i ) is the covariance kernel, and the Gaussian kernel is an indicator measuring the proximity of points x i and y i . The joint distribution of the training set {(x i , y i ), i = 1, 2,..., n} and the predicted test output (x * , y * ) is:

[0035]

[0036] Predicted covariance:

[0037] Δ 2 =K(x * , x * ) - K(x * , X)(K(X, X)+σ 2 I) -1 K(X, x * )

[0038] y * represents the prediction result, x is the input data, ∈ i is the observation error of the input data, and the observation error follows a Gaussian distribution where σ is the variance.

[0039] A power battery and an energy storage battery capacity prediction system, comprising a memory and a processor;

[0040] Wherein, the memory stores a computer program, and when the program is executed by the processor, it can implement a method for predicting the capacity of a power battery and an energy storage battery.

[0041] Advantages of the present invention: The algorithm of the present invention has extremely high accuracy and extremely fast speed, which can provide a solid foundation for miniaturized and highly reliable battery health measurement equipment. At the same time, the HI based on the geometric features related to the SEI film has the highest accuracy in predicting the battery health state, and the MAE of the battery health state estimation can reach within 1%. Description of the Drawings

[0042] Figure 1 It is the original data graph of the battery EIS;

[0043] Figure 2 It is the fitting semi-circle data graph of the EIS specific frequency band;

[0044] Figure 3 It is the schematic diagram of the fitting semi-circle;

[0045] Figure 4 It is the schematic diagram of the two-layer GPR;

[0046] Figure 5 It is the Nyquist graph of the EIS data of the 35C02 sequence;

[0047] Figure 6 It is the HI1 feature extracted from the EIS data of the 35C02 sequence;

[0048] Figure 7 It is the HI2 feature extracted from the EIS data of the 35C02 sequence;

[0049] Figure 8 It is the HI3 feature extracted from the EIS data of the 35C02 sequence;

[0050] Figure 9 It is the prediction graph of the battery health state of different HIs of the 35C02 sequence;

[0051] Figure 10 It is the health state estimation graph of 35C01;

[0052] Figure 11 It is the health state estimation error graph of 35C01;

[0053] Figure 12 It is the health state estimation graph of 35C00;

[0054] Figure 13It is a health state estimation error graph for 35C02. Detailed implementation manners

[0055] The present invention will be further described below with reference to the accompanying drawings.

[0056] Please refer to Figures 1 to 4 , the present invention provides an embodiment: a method for predicting the health state of a power battery and an energy storage battery, characterized in that: the method comprises the following steps:

[0057] Step S1: Obtain the original EIS data through an open-source data set, and extract the center position, radius and arc length of the arc of the mid-high frequency Nyquist diagram, as well as the length of the Nyquist diagram in the low-frequency band as health features, and obtain a training set; the original EIS data of the battery is as Figure 1 , Figure 2 shown. Research shows that during the internal electrochemical reaction process of the battery, electrolyte substances precipitate to form an SEI film, and the SEI film continuously grows with the increase of the charge and discharge times, thereby increasing the resistance of lithium ion diffusion, manifested as an increase in battery resistance, resulting in the attenuation of battery health.

[0058] The process of SEI film growth is manifested as the first semi-circle of the Nyquist diagram of EIS data, as Figure 3 shown. As the battery ages, the semi-circle shape, semi-circle position and degradation process shape change continuously. Capturing this change through geometric features can reflect the attenuation of the battery health state. As Figure 2 shown, as the battery ages, the position of the fitted semi-circle changes continuously, and the radius of the semi-circle shows an increasing trend. As Figure 3 shown, this patent uses the product of the fitted semi-circle radius and the center position formula to describe the SEI film thickness; uses the intersecting arc length of the frequency band fitted semi-circle and the EIS data to describe the time required for SEI film growth; uses the shape length of the EIS Nyquist diagram of the battery electrochemical diffusion process to describe the influence of the SEI film on the battery performance.

[0059] Step S2: Construct a DGPR model; input the training set into DGPR for training, select a kernel function, train to obtain hyperparameters, and establish an accurate battery health state prediction model; this patent uses the machine learning Gaussian process (GPML) toolbox to implement the GPR model, and this toolbox has a zero-mean function and a covariance function of the Matern class ARD (Automatic relevance determination, ARD):

[0060]

[0061] where σ m represents the length scale of feature m, and σ f is the signal standard deviation; these hyperparameters are obtained by maximizing the marginal likelihood.

[0062] Deep Gaussian Process Regression (DGPR) is a hierarchical generalization of GPR. It combines the uncertainty calculated by GPR with a multi-layer model and high flexibility, and can learn complex functional relationships at multiple scales, thus obtaining stronger fitting ability. A two-layer DGPR is shown as Figure 4 shown. This patent uses g i to represent the hidden function value of the i-th input.

[0063] Step S3: Input the health features extracted from the test set into the trained health state estimation model, and predict the battery health state according to the hyperparameters obtained by training.

[0064] Please continue to refer to Figures 1 to 4 shown. In an embodiment of the present invention, the health feature set X = [x1,..., x n T and the battery health state set Y = [y1,..., y n T corresponding to the health features are defined. The training set in step S1 is {(x i , y i ), i = 1, 2,..., n}. The input value x i = [HI1, HI2, HI3], including the weighted radius calculated by fitting the real and imaginary parts of the impedance spectrum collected from 20KHz to 45Hz in the current cycle to the center coordinate position and radius of the semi-circle, as well as the length of the semi-circular arc and the length of the shape of the diffusion process EIS Nyquist plot. The output value y i is the battery health state corresponding to the health features. The real and imaginary parts of the EIS data set and the EIS sequence data in the open source data set are normalized using the mean and standard deviation of the data in the training set. The open source data set contains EIS data measured under different temperatures and different batteries. A sequence data represents the data set measured by a certain battery at a certain temperature. The data set contains the impedance (EIS) measured in the current cycle and the battery capacity. Only the impedance data is normalized during training.

[0065] Please continue to refer to Figures 1 to 4 shown. In an embodiment of the present invention, the acquisition method of the health features in step S1 is:

[0066] Using the real and imaginary part data EIS of the EIS data in the frequency range of [20kHz to 45Hz], a total of 60 frequencies 1-26 ​​Calculate the center position and radius of the first semi - circle in the Nyquist plot, as well as the polar coordinates of the center. The Nyquist plot of the battery EIS is close to a semi - circle in the medium - high frequency range. Therefore, the EIS characteristics in the medium - high frequency range can be described as:

[0067] (Z′ - X) 2 +(Z″ - Y)=R 2

[0068] Among them, EIS 1-26 represents the data corresponding to the 1st to 26th frequencies (from large to small) among 60 frequency EIS data of a certain sequence in the data set. Z′ and Z″ are the real - part and imaginary - part data of EIS; X and Y are the center positions of the Nyquist semi - circle, and R is the radius of the semi - circle. For each data point, the geometric distance from it to the center is:

[0069]

[0070] The objective function for fitting the semi - circle is:

[0071]

[0072] Use the gradient - descent method to obtain X, Y, and R. The formula for health feature 1 is:

[0073]

[0074] Obtain the included angle θ between the two end - points of the selected frequency segment and the center of the semi - circle, and then obtain the arc length corresponding to the fitted semi - circle of the selected frequency segment. The formula for health feature 2 is:

[0075] HI2=R×θ

[0076] Obtain the EIS 27-60 The EIS data corresponding to the minimum value of the imaginary part is the starting point (S) of the battery electrochemical diffusion process. The EIS 60 is used as the end - point (E). The formula for health feature 3 is:

[0077]

[0078] Among them, S z′ represents the real part of the EIS corresponding to the starting point, and S z″ is the imaginary part of the starting point; the EIS 27-60 represents the data corresponding to the 27th to 60th frequencies (from large to small) among 60 frequency EIS data of a certain sequence in the data set. The EIS 60 represents the data of the 60th frequency point of a certain sequence of EIS data in the data set; E Z' represents the real part of the end - point of the diffusion process, and E Z” represents the imaginary part of the end - point of the diffusion process.

[0079] Please continue to refer to Figures 1 to 4 As shown, in one embodiment of the present invention, step S2 is as follows:

[0080] Given a training set τ = {(x i and output y i ), i = 1, 2, 3,..., n} consisting of n pairs of input x i and output y i ), calculate the predictive distribution of the unknown observed value y * at the test metric x * . The GPR assumes that any set of function values f(x) = [f(x1), …, f(x n )] follows a multivariate Gaussian distribution, that is:

[0081]

[0082] where K is a matrix whose elements are K ij = k(x i , x j ), and k(·, ·) is the kernel function;

[0083] Input the training set into DGPR for training, select MaternARD 3 / 5 as the kernel function, and train to obtain the hyperparameters hyp = [lik, cov1, cov2,..., cov n ,

[0084] lik represents the log standard deviation of the observation noise, which describes the magnitude of the noise in the data;

[0085] cov1 describes the range of variation of the input features, indicating how close two data points need to be to be considered relevant;

[0086] cov2 - n describes the magnitude of the output value of the kernel function, that is, the overall variation magnitude of the data; Use 35C01 / 35C02 as the test set, and perform battery health state prediction according to the trained hyperparameters. EIS data can provide details about the internal electrochemical process of the battery. Extract battery health features from these details, use GPR to map the feature data to a higher-dimensional feature space, capture more complex patterns and relationships, and thus establish an accurate battery health state prediction model.

[0087] Please continue to refer to Figures 1 to 4 As shown, in one embodiment of the present invention, step S3 is as follows:

[0088] GPR performs non-parametric regression using a Gaussian process, assuming is independent and identically distributed Gaussian noise, and the output f = (f(x1),..., f(x n)) is modeled as a Gaussian random field

[0089] where K ij = k(x i , y i ) is the covariance kernel, and the Gaussian kernel is an indicator for measuring the proximity of points x i and y i . The joint distribution of the training set {(x i , y i ), i = 1, 2,..., n} and the predicted test output (x * , y * ) is:

[0090]

[0091] Predicted covariance:

[0092] Δ 2 = K(x * , x * ) - K(x * , X)(K(X, X) + σ 2 I) -1 K(X, x * )

[0093] y * represents the prediction result, x is the input data, ∈ i is the observation error of the input data, and the observation error follows a Gaussian distribution where σ is the variance.

[0094] Please refer to Figures 1 to 4 , the present invention provides another embodiment: a power battery and an energy storage battery capacity prediction system, including a memory and a processor;

[0095] wherein, the memory stores a computer program, and when the program is executed by the processor, it can implement a method for predicting the capacity of a power battery and an energy storage battery. Specific embodiment:

[0097] The experimental data of this embodiment are from the Cavendish Laboratory of the University of Cambridge. The test sample is a commercially available LiCoO2 / graphite battery. After charge-discharge cycling, 12 battery samples were measured at working temperatures of 25 °C (25C01 - 25C08), 35 °C (35C01 and 35C02), and 45 °C (45C01 and 45C02). Under various operating conditions at nine different stages (denoted as state 1, … 9) (1 represents measurement before charging, 5 represents measurement after full charge and standing, 9 represents measurement after full discharge and standing, and 2 / 3 / 4 / 6 / 7 / 8 represent measurement during charge-discharge), EIS measurements were carried out at a total of 60 frequencies in the frequency range of 0.02 Hz to 20 kHz. Then the battery capacity was measured after an odd number of cycles. This patent uses EIS and capacity data calculated by cycle. Since it is more valuable to measure the actual health state of the battery after full charge and standing, this patent uses the sequence data in state 5. Taking battery 35C02 as an example, the relationship between EIS and the number of cycles is as Figure 5 shown.

[0098] (1) Using the real and imaginary part data of the EIS data in the frequency range of [20 kHz ~ 45 Hz] (EIS 1-26 ), calculate the center position and radius of the first semi-circle of the Nyquist plot, as well as the polar coordinates of the center. As Figure 3 shown, the Nyquist plot of the battery EIS in the medium and high frequency range is close to a semi-circle. Therefore, the EIS characteristics in the medium and high frequency range can be described as:

[0099] (Z′—X) 2 +(Z″ - Y) = R 2

[0100] where Z′ and Z″ are the real and imaginary part data of the EIS; X and Y are the center positions of the Nyquist semi-circle, and R is the radius of the semi-circle. For each data point, the geometric distance from it to the center is:

[0101]

[0102] The objective function for fitting the semi-circle is:

[0103]

[0104] Use the gradient descent method to calculate X, Y, and R. The calculation formula for health feature 1 is:

[0105]

[0106] Calculate the included angle θ between the two points at the beginning and end of the selected frequency range and the center of the semi-circle, and then calculate the arc length corresponding to the fitted semi-circle in the selected frequency range. The calculation formula for health feature 2 is:

[0107] HI2 = R×θ

[0108] Calculating EIS 27-60 The EIS data corresponding to the minimum value of the imaginary part in 27-60 is the starting point (S) of the battery electrochemical diffusion process, and the EIS 60 is used as the end point (E). The calculation formula for health feature 3 is:

[0109]

[0110] ; where S z′ represents the real part of the EIS corresponding to the starting point, and S z″ is the imaginary part of the starting point.

[0111] The schematic diagram of the health feature calculation result is as shown in Figures 5 - 8 shown.

[0112] (3) Based on the GPML toolbox, calculate {HI1, HI2, HI3} of the EIS data of {25C01, 25C02, 25C03, 25C04, 35C01, 35C02, 45C01, 45C02}, and combine the true value SOH as the training set to input DGPR for training. Select MaternARD 3 / 5 as the kernel function. The trained hyperparameters are hyp = [lik, cov1, cov2,..., cov n ,,

[0113] lik represents the logarithmic standard deviation of the observation noise, which describes the magnitude of the noise in the data;

[0114] cov1 describes the variation range of the input features, indicating how close two data points need to be considered relevant;

[0115] cov2 - n describes the amplitude of the kernel function output value, that is, the overall variation amplitude of the data. Use 35C01 / 35C02 as the test set, and predict the battery health state according to the trained hyperparameters.

[0116] (4) To verify the performance of the method of this patent, extract different health features (HI6, RXY) of previous patents as comparison features. The experimental steps are the same as those in (3) above. Among them, the HI6 health feature refers to the imaginary part extreme value and the corresponding real part and amplitude between 20 kHz and 0.1 Hz, the imaginary part and amplitude corresponding to 0.1 Hz, and the phase extreme value; RXY is the data of the center and radius of the semi - circle fitted from the real part and imaginary part data in the frequency range of [20 kHz to 45 Hz]. Use the Mean Absolute Error (MAE) as the evaluation index, and its calculation formula is: The result is as shown in Figures 9 to 13As shown in Table 1, it can be seen that the HI based on the geometric features related to the SEI film proposed in this paper has the highest accuracy in predicting the battery health state, and the MAE of the battery health state estimation can reach within 1%. These results indicate that the proposed HI has good performance.

[0117] 35c01 35c02 This patent 0.022 0.0097 RXY 0.0669 0.0144 HI6 0.0867 0.0406

[0118] Table 1 MAE of different HI battery health state predictions for the 35° sequence

[0119] The above are only the preferred embodiments of the present invention and should not be construed as limitations on this application. All equivalent changes and modifications made in accordance with the scope of the patent application of the present invention shall fall within the scope of the present invention.

Claims

1. A method for predicting the health status of a power battery and an energy storage battery, characterized in that: The method comprises the following steps: Step S1: Obtain the original EIS data through the open source data set, and extract the center position, radius and arc length of the mid- and high-frequency Nyquist diagram arc, as well as the length of the low-frequency Nyquist diagram as health features to obtain the training set; Step S2: construct a DGPR model; input the training set into the DGPR for training, select the kernel function, train to obtain hyperparameters, and establish an accurate battery health status prediction model; Step S3: Input the health features extracted from the test set into the trained health status estimation model, and predict the battery health status according to the hyperparameters obtained through training.

2. A method for predicting the capacity of a power battery and an energy storage battery according to claim 1, characterized in that: Define the health feature set X = [x1, ..., x n ] T The battery health status set Y = [y1, ..., y n ] T The training set of step S1 is {(x i ,y i ), i = 1, 2, ..., n}, input value, x i = [HI1, HI2, HI3], including the weighted radius calculated by fitting the coordinate position and radius of the center of the semicircle using the real and imaginary parts of the impedance spectrum collected from 20KHz to 45Hz in the current cycle, as well as the length of the semicircle arc and the shape length of the diffusion process EIS Nyquist diagram, and the output value y i It is the battery health status corresponding to the health feature. The real and imaginary data sets of the EIS data and the EIS sequence data in the open source data set are normalized using the mean and standard deviation of the training set data.

3. The method for predicting the capacity of a power battery and an energy storage battery according to claim 1, characterized in that: The health characteristics of step S1 are obtained in the following manner: Using EIS data in the frequency range [20kHz~45Hz], a total of 60 frequencies of real and imaginary data EIS 1-26 Calculate the center position and radius of the first semicircle of the Nyquist diagram, as well as the polar coordinates of the center. The Nyquist diagram of the battery EIS in the medium and high frequency range is close to a semicircle, so the EIS characteristics in the medium and high frequency range can be described as: (Z′-X) 2 +(Z″-Y)=R 2 Among them, EIS 1-26 It represents the data corresponding to the 1st to 26th frequencies in the 60 frequencies (from large to small) of a certain sequence of EIS data in the data set. Z′ and Z″ are the real and imaginary data of EIS. X and Y are the center positions of the Nyquist semicircle, and R is the radius of the semicircle. For each data point, its geometric distance to the center is: The objective function of fitting a semicircle is: Use gradient descent to obtain X, Y, R. The formula for health feature 1 is: The angle θ between the first and last points of the selected frequency segment and the center of the semicircle is obtained, so as to obtain the arc length corresponding to the fitted semicircle of the selected frequency segment. The formula of health feature 2 is: HI2=R×θ Obtaining an EIS 27-60 The EIS data corresponding to the minimum value of the imaginary part is the starting point S of the battery electrochemical diffusion process. 60 As the endpoint E, the formula for health characteristic 3 is: Among them, S Z′ represents the real part of the EIS corresponding to the starting point, S Z″ Imaginary part of the starting point; EIS 27-60 Indicates the data corresponding to the 27th to 60th frequency of the EIS data of a certain sequence in the data set (from large to small). EIS 60 Represents the data of the 60th frequency point of a sequence of EIS data in the data set; E Z' represents the real part of the diffusion process end point, E Z” represents the imaginary part of the end point of the diffusion process.

4. A method for predicting the capacity of a power battery and an energy storage battery according to claim 1, characterized in that: The step S2 is: Given an input x consisting of n pairs i and output y i The training set τ = {(x i ,y i ), i = 1, 2, 3, ..., n}, calculate the test index x * The unknown observation y * The GPR assumes that any set of function values ​​f(x) = [f(x1),…,f(x n )] as a multivariate Gaussian distribution, that is: Where K is a matrix whose elements are K ij = k(x i , x j ), k(·,·) is the kernel function; Input the training set into DGPR for training and select MaternARD 3 / 5 As the kernel function, the training obtains the hyperparameter hyp=[lik, cov1, cov2, ..., cov n ], lik represents the logarithmic standard deviation of the observed noise, which describes the magnitude of the noise in the data; cov1 describes the range of variation of the input features, indicating how close two data points need to be to be considered related; cov2-n describes the amplitude of the kernel function output value, that is, the amplitude of the overall data change; 35C01 / 35C02 is used as the test set, and the battery health status is predicted based on the hyperparameters obtained through training.

5. The method for predicting the capacity of a power battery and an energy storage battery according to claim 1, characterized in that: The step S3 is: GPR performs nonparametric regression using Gaussian processes, assuming is independent and identically distributed Gaussian noise, output f = (f(x1), ..., f(x n )) is modeled as a Gaussian random field Where K ij = k(x i ,y i ) is the covariance kernel, and the Gaussian kernel is a measure of the point x i With y i The index of closeness, the training set {(x i ,y i ), i = 1, 2, ..., n} and predict the test output (x * ,y * ) is: Prediction covariance: Δ 2 =K(x * ,x * )-K(x * ,X)(K(X,X)+σ 2 I) -1 K(X,x * ) y * Represents the prediction result, x is the input data, ∈ i is the observation error of the input data, and the observation error follows a Gaussian distribution where σ is the variance.

6. A power battery and energy storage battery capacity prediction system, characterized in that: including memory and processor; Wherein, the memory stores a computer program, and when the program is executed by the processor, it can implement a power battery and energy storage battery capacity prediction method as described in any one of claims 1 to 5.

Images