Lithium battery state estimation method based on relaxation time scale modeling

Abstract

The invention discloses a lithium battery state estimation method based on relaxation time scale modeling, and the method comprises the steps: 1, carrying out a sudden pulse charging test on a lithium battery, collecting voltage and current data, and calculating the open-circuit voltage, direct-current internal resistance and a time-domain impedance sequence of the lithium battery; and 2, converting the time domain impedance sequence of the lithium battery into a frequency domain impedance sequence after discrete Fourier transform, establishing a fitting equation set of a distribution function of the relaxation time scale DRT of the lithium battery and the frequency domain impedance sequence, and solving by adopting a regularized least square method. And 3, taking the solved DRT model distribution function as the input of a neural network, taking the state parameters of the lithium battery as the output of the neural network, and establishing a prediction model of the state parameters of the lithium battery. According to the method, EIS test required by traditional relaxation time scale modeling is avoided, the control complexity and design cost of test acquisition equipment are effectively reduced, and the method can be effectively applied to occasions such as battery life prediction, battery classification and screening, battery echelon utilization and the like.

Description

Technical Field

[0001] The invention relates to the field of lithium battery recycling and BMS application, and in particular to a lithium battery state estimation method based on relaxation time scale modeling. Background Art

[0002] Lithium-ion batteries are considered to be one of the most promising on-board energy sources for electric vehicles (EVs) due to their high energy density, high power capability, long cycle life, and light weight. The pursuit of safer, more reliable, more efficient, and longer-lasting electrochemical energy storage and conversion technologies cannot be separated from the support of lithium battery state parameter (including state of charge SOC and state of health SOH) estimation technology. Existing methods for estimating lithium battery state parameters are mainly based on modeling methods, including electrochemical models, thermodynamic models, coupling models, equivalent circuit models, etc. Among them, the equivalent circuit model law has the advantage of not requiring in-depth analysis of the electrochemical reactions inside the battery. It describes the battery's open circuit voltage, DC internal resistance, and polarization internal resistance through circuits to achieve the characterization of the battery's external characteristics.

[0003] However, the existing equivalent circuit model method still has many defects and shortcomings, such as the need for a priori assumptions. For example, the result of the interaction between the internal and external parameters of the lithium battery system will be presented on a limited bandwidth, causing its characteristic peaks to overlap or the characteristic time constants to be not very different and difficult to separate. Therefore, modeling methods based on the relaxation time scale, such as the DRT method, have been proposed to achieve high-precision modeling. However, the traditional DRT model parameter identification method often requires EIS testing, which has the disadvantages of difficulty in testing and long testing time (relative to the method of the present invention) in real applications (large batteries, large modules, BMS, charging piles, etc.). Summary of the invention

[0004] The present invention aims to avoid the shortcomings of the above-mentioned prior art and provide a lithium battery state estimation method based on relaxation time scale modeling, so as to realize parameter identification of the lithium battery DRT model through time domain pulse testing, thereby realizing convenient and rapid estimation of the lithium battery state parameters.

[0005] The present invention adopts the following technical solutions to solve the technical problems:

[0006] The present invention provides a lithium battery state estimation method based on relaxation time scale modeling, which comprises the following steps:

[0007] Step S1: The lithium battery to be tested is left to stand for a period of time, and after the lithium battery is left to stand, it is connected to a high-speed voltage and current sensor, and then a sudden charging pulse test is performed on the lithium battery to obtain a relative sampling time series. , current sampling sequence and voltage sampling sequence ,in, Indicates the nth moment of the sudden charging pulse, represents the mth moment when the charging pulse ends, and Respectively represent the xth moment The voltage and current data below, where ;

[0008] Step S2: The average value of the open circuit voltage of the lithium battery is , using the nth moment Voltage data under , the n+1th moment Voltage data under And the applied current pulse value I calculates the DC internal resistance of the lithium battery ;

[0009] Step S3: construct the relaxation time distribution model DRT of the lithium battery and convert the voltage sampling sequence As the port output voltage of the relaxation time distribution model DRT, the current sampling sequence As the main circuit current of the relaxation time distribution model DRT, the relaxation time distribution model DRT of the lithium battery is constructed using formula (1) at the xth moment The voltage data, current data and impedance equation are as follows:

[0010] (1)

[0011] In formula (1), Indicates the infinite polarization process in series in the relaxation time distribution model DRT at the xth moment The impedance under

[0012] Step S4, calculating the discrete impedance sequence measurement value of the lithium battery in the frequency domain;

[0013] Step S4.1, take the nth moment and the mth moment The voltage and current data between the sudden pulse and the pulse end time are calculated, and the impedance discrete sequence of the relaxation time distribution model DRT of the lithium battery in the time domain t is calculated. ,in, The relaxation time distribution model DRT is represented in Moment The impedance under ; In the relaxation time distribution model DRT, the infinite polarization process is connected in series. Moment The impedance under ;

[0014] Step S4.2: Using discrete Fourier transform Transform and write each frequency domain component after transformation into complex form, so as to obtain the impedance discrete sequence of lithium battery in frequency domain f ,in, The relaxation time distribution model DRT is represented in Frequency The impedance under , express The real part of express The imaginary part of represents the imaginary unit, ;

[0015] Step S5: Fitting and solving the distribution function of the relaxation time distribution model DRT ;in, represents the time constant;

[0016] Step S5.1, using formula (2) to establish the relaxation time distribution model DRT in the first Frequency The distribution function under The fitting equation is:

[0017] (2)

[0018] In formula (2), represents the differential symbol;

[0019] Step S5.2: Use the radial basis function RBF with a Gaussian kernel function to transform the distribution function at each frequency Discretize the DRT function Converted into the sum of a series of basis functions with different weights, and the distribution function shown in formula () is obtained The fitting equations are:

[0020] (3)

[0021] In formula (3), and yes The real and imaginary parts of and The dimension is The real and imaginary matrices of The length of , and N = mn, M is The number of basis functions after discretization; is the weight vector of the M basis functions to be solved;

[0022] Step S5.3: Solve the fitting equations using the least squares method with regularization term to obtain the optimal weight vector , and then get the DRT distribution function ;

[0023] Step S6: DRT distribution function The state of charge or health state of the lithium battery is used as the input of the neural network and the output of the neural network, so as to train the neural network and obtain a state of charge prediction model or a health state prediction model of the lithium battery state parameters.

[0024] The electronic device of the present invention includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the lithium battery state estimation method, and the processor is configured to execute the program stored in the memory.

[0025] The present invention provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and the computer program executes the steps of the lithium battery state estimation method when the computer program is executed by a processor.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0027] 1. Compared with the traditional Rint model, Thevenin model, DP model and PNGV, the relaxation time scale model established by the method of the present invention has higher model accuracy, is closer to the actual battery mechanism process and polarization reaction process, and has better model interpretability, thereby providing richer feature selection in the state estimation algorithm of lithium batteries.

[0028] 2. The method of the present invention solves the problem of often needing to consider a priori assumptions when using an equivalent circuit model, that is, it is necessary to know or pre-assume in advance that there are several characteristic time constants in the impedance spectrum, but this is often difficult to determine. The DRT function under the relaxation time scale can more accurately fit different types of lithium batteries without trial and error, and has a higher fitting accuracy for the impedance spectrum and dynamic characteristics of the lithium battery, thereby effectively improving the accuracy of the battery state estimation algorithm.

[0029] 3. The model parameter identification test condition established by the method of the present invention is different from the traditional EIS test technology. Through short-term sudden charging pulse testing, the complexity of the working condition and the control cost are reduced, providing a practical solution for practical applications.

[0030] 4. The method of the present invention is to model and identify the parameters of lithium batteries on the relaxation time scale under a test condition that is easy to implement and fast, and to extract features through a neural network, thereby realizing a convenient and fast estimation of the state parameters of the lithium battery. It has high operability and practical value, and can be applied to occasions such as lithium battery recycling and electric vehicle BMS design. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is a model diagram of the relaxation time scale of lithium batteries under typical DRT analysis technology;

[0032] Figure 2a is an equivalent response circuit diagram of a lithium battery under a pulse current in the present invention;

[0033] Figure 2b It is a current and voltage curve diagram collected by the lithium battery under pulse current in the present invention;

[0034] Figure 3 It is an analysis diagram of the piecewise linear interpolation method involved in the present invention;

[0035] Figure 4 is the relaxation time distribution function fitted in the present invention Schematic diagram of

[0036] Figure 5 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0037] In this embodiment, a lithium battery state estimation method based on relaxation time scale modeling is mainly to fit the lithium battery relaxation time distribution (DRT) model function by short-time pulse test data. , and based on the DRT function obtained by fitting , use neural network algorithm to extract features, so as to establish a prediction model of lithium battery state parameters. Specifically, Figure 5 As shown, the method comprises the following steps:

[0038] Step S1, let the lithium battery to be tested stand for a period of time, ensuring that the standing time is more than 1 hour; connect the lithium battery after standing to the high-speed voltage and current sensor, and perform a sudden charging pulse test on the lithium battery. The test condition is set as follows: after standing for 10 seconds, a current excitation with an amplitude of I=2C is suddenly applied, the time ends after 60 seconds, and the test ends after standing for 10 seconds, thereby obtaining a sampling frequency of The voltage and current data are sampled as follows: Figure 2b As shown in the figure, the data acquisition adopts relative time, that is, the timing starts from 10s before the sudden pulse, including the relative sampling time series , current sampling sequence and voltage sampling sequence ,in, Indicates the nth moment, at which a charging pulse is suddenly added. represents the mth moment, at which the charging pulse ends. and Indicates Moment Voltage and current sampling data under time, , and is an integer;

[0039] Step S2: sampling the current sampling sequence data under the sampling time sequence and voltage sampling sequence data Calculate and take The mean value is the open circuit voltage of lithium battery , the calculation formula is shown in formula (1):

[0040] (1)

[0041] In formula (1), since the test time is short and the test impact on the battery is small, it can be considered that the state of the lithium battery is not changed, and the test process can be Considered to be a constant.

[0042] Then according to Figure 2a With formula (2), using , The DC internal resistance of the lithium battery can be calculated by (ohmic internal resistance), where I is the amplitude of the applied current pulse, which the present invention considers to be approximately constant throughout the process:

[0043] (2)

[0044] In formula (2), since the test time is short and the test effect on the battery is small, it can be considered that the state of the lithium battery is not changed. Therefore, the present invention considers It remains approximately a constant value throughout the process.

[0045] Step S3: Figure 1 As shown, a model based on relaxation time distribution (DRT) is established to describe the internal impedance of the lithium battery, which is used as the lithium battery model used in the present invention. In the illustrated battery impedance model, It represents the ohmic impedance, i.e. the DC impedance part, and uses an RC parallel structure that tends to infinity to represent the impedance of the polarization process. Its related parameters are the unknown quantities that need to be identified.

[0046] When applying a sudden charging pulse test to a lithium battery, the lithium battery charging circuit is as follows Figure 2aAs shown in the figure, the relaxation time distribution model of lithium battery is constructed in the Moment The voltage data, current data and impedance equation under are shown in formula (3):

[0047] (3)

[0048] In formula (3), express The impedance value of the series part of the infinite polarization process in the DRT model at the time, and express The voltage and current sampling data at the moment,

[0049] Step S4: Calculate the discrete impedance sequence measurement value of the lithium battery in the frequency domain.

[0050] Step S4.1, rewrite the equation into equation (4):

[0051] (4)

[0052] During the lithium battery pulse test, no excitation signal is applied to the lithium battery before the pulse current starts and after the pulse ends. Therefore, when calculating the lithium battery time domain impedance sequence, only the pulse start moment is taken. and the pulse ends The discrete impedance sequence of the battery changing with time is calculated using formula (4): .

[0053] Step S4.2: For the discrete time sequence Using the discrete Fourier transform shown in equation (5), we can obtain the frequency domain impedance series: :

[0054] (5)

[0055] Mode middle, is the frequency domain impedance series No. Frequency The impedance component below: is the time domain impedance series in The impedance at the moment, and yes The voltage and current sampling data at the moment, ; is the discrete impedance sequence in the time domain The length of ; is an imaginary unit, satisfying . Through Euler's formula , the frequency domain discrete sequence Written in complex form as shown in formula (6):

[0056] (6)

[0057] Mode middle, Indicates the frequency index, which means starting from 1, the frequency index calculated by discrete Fourier transform (DFT) After discrete Fourier transform, the frequency point number The frequency corresponding to each point, that is, the analog domain frequency is given by formula (7):

[0058] (7)

[0059] Mode middle, is the signal sampling frequency, is the discrete impedance sequence in the time domain According to formula (8), the length of the discrete Fourier transform can be obtained The simulated angular frequency corresponding to the point :

[0060] (8)

[0061] After calculations from equations (5) to (8), the impedance discrete sequence of the lithium battery in the frequency domain f is obtained: ,in, The relaxation time distribution model DRT is represented in Frequency The impedance under , express The real part of express The imaginary part of ;

[0062] Step S5: discrete impedance sequence in the frequency domain f obtained in step S3 Fitting to obtain the DRT model distribution function ;in, represents the time constant;

[0063] Step S5.1, apply Fourier transform to both ends of equation (4) to obtain the frequency domain form of the voltage equation of the lithium battery impedance model as shown in equation (9):

[0064] (9)

[0065] In formula (9), This is the impedance frequency domain expression of the infinite polarization process part of the DRT model. is the impulse function, defined as The values ​​of all points outside are 0, but the integral is 1. Based on existing theoretical knowledge, the DRT method assumes that the response of the electrochemical system under study is obtained from the relaxation distribution. The impedance can be written as (10):

[0066] (10)

[0067] Substituting formula (10) into formula (9), we can get formula (11):

[0068] (11)

[0069] In formula (11): The frequency domain impedance sequence of the lithium battery is obtained after Fourier transform of the discrete impedance sequence in the time domain.

[0070] According to formula (6), the frequency domain impedance series , using formula (12), we can get the DRT distribution function at each frequency The fitting equation is:

[0071] (12)

[0072] In the fitting process equation (12), the frequency domain impedance component The corresponding frequency is equal to , Represents the differential symbol.

[0073] Step S5.2: Fitting the DRT function Before that, you need to first Discretization is done by It is realized by decomposing into a series of basis functions. There are currently three common basis functions: Dirac function, piecewise linear function (PWL) and radial basis function (RBF). Here, RBF with Gaussian function is used as the discretization basis. The expression of RBF with Gaussian function as kernel function is as shown in formula (13):

[0074] (13)

[0075] In formula (13), is the mean center given by the corresponding basis function. is the shape factor describing the RBF distribution, is the height of the curve peak, is the standard deviation, such as Figure 3As shown, each basis function multiplied by the corresponding weight and then summed can approximate the DRT function In the DRT function Applying the RBF basis, the expression of DRT can be rewritten as formula (14):

[0076] (14)

[0077] In formula (14), is the number of basis functions, is the weight of each basis function. Therefore, the DRT can be obtained from the discrete frequency domain impedance data obtained by Equation (6) by calculating the weight of the basis function. The RBF-based fitting can achieve fast numerical convergence and eliminate the edge oscillation effect.

[0078] DRT function After discretization, the DRT model distribution function can be obtained using formula (15): In frequency The following linear fitting equations are:

[0079] (15)

[0080] Equation (15) holds true at every frequency, so we can get the DRT function The complete fitting equation of is written in the matrix form shown in formula (16):

[0081] (16)

[0082] In formula (16), and Is the size of matrix, is the discrete frequency domain impedance series Length, yes The number of basis functions after discretization, That is, the weight of each basis function that needs to be solved. , we can get the DRT function , that is, the parameter identification of the lithium battery relaxation time scale model is completed. It is calculated by the regularized least squares method.

[0083] Step S5.3, the solution of equation (16) is generally obtained by the least squares method with regularization term as shown in equation (17):

[0084] (17)

[0085] Mode middle, is the regularization term, is the regularization parameter. Common regularization terms are L0 norm, L1 norm and L2 norm. Solve the equation Transformed into minimizing the objective function , and minimize the objective function It can be regarded as a Tikhonov regularization problem. When writing a computer program, by properly choosing the regularization parameter , we can get the optimal weight of the basis function , thus obtaining the distribution function under the lithium battery relaxation time scale model .

[0086] Step S6: DRT distribution function obtained in step S5 like Figure 4 As shown, it is used as the input of the neural network, the characteristic parameters are extracted through the neural network, and the state parameters of the lithium battery (including the state of charge SOC and the state of health SOH) are used as the output of the neural network, so as to train the neural network and obtain the prediction model of the lithium battery state parameters.

[0087] In this embodiment, an electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the above method, and the processor is configured to execute the program stored in the memory.

[0088] In this embodiment, a computer-readable storage medium stores a computer program on the computer-readable storage medium, and the computer program executes the steps of the above method when executed by a processor.

[0089] In summary, the method of the present invention first performs a charging pulse test on the lithium battery to obtain time-domain voltage and current data, and calculates the time-domain impedance discrete sequence of the lithium battery based on the test data. Then, in order to establish the fitting equation of the DRT function, the discrete Fourier transform is used to convert the time-domain impedance discrete sequence of the lithium battery into a frequency-domain discrete sequence. Each element in the obtained frequency-domain discrete sequence corresponds to a frequency, representing the impedance of the lithium battery at this frequency, thereby establishing the theoretical calculation value of the DRT model impedance (through the function represents) has a one-to-one correspondence with the actual value. is a nonlinear function and cannot be solved directly, so it needs to be discretized to obtain The complete fitting equation system of , so that the function can be solved by regularized least squares method The obtained function It can be used to characterize the changes in the parameters of the DRT model of lithium batteries. The neural network can be used to extract the strong correlation characteristics of the lithium battery state of charge SOC or health state SOH, thereby establishing a prediction model for the lithium battery SOC or SOH. The method of the present invention uses a fast and easy-to-implement pulse test to replace the traditional inconvenient EIS test to fit the distribution function of the lithium battery DRT model. In addition, since the DRT model used has a high fitting accuracy for the lithium battery impedance spectrum and dynamic behavior, it effectively reduces the complexity and cost of the test conditions in the lithium battery state estimation process while maintaining a high accuracy of lithium battery state parameter estimation, and has strong application value. It can be applied to lithium battery cascade utilization and electric vehicle BMS design.

Claims

1. A lithium battery state estimation method based on relaxation time scale modeling, characterized in that: The steps include: Step S1: The lithium battery to be tested is left to stand for a period of time, and after the lithium battery is left to stand, it is connected to a high-speed voltage and current sensor, and then a sudden charging pulse test is performed on the lithium battery to obtain a relative sampling time series. , current sampling sequence and voltage sampling sequence ,in, Indicates the nth moment of the sudden charging pulse, represents the mth moment when the charging pulse ends, and Respectively represent the xth moment The voltage and current data below, where ; Step S2: The average value of the open circuit voltage of the lithium battery is , using the nth moment Voltage data under , the n+1th moment Voltage data under And the applied current pulse value I calculates the DC internal resistance of the lithium battery ; Step S3: construct the relaxation time distribution model DRT of the lithium battery and convert the voltage sampling sequence As the port output voltage of the relaxation time distribution model DRT, the current sampling sequence As the main circuit current of the relaxation time distribution model DRT, the relaxation time distribution model DRT of the lithium battery is constructed using formula (1) at the xth moment The voltage data, current data and impedance equation are as follows: (1) In formula (1), Indicates the infinite polarization process in series in the relaxation time distribution model DRT at the xth moment The impedance under Step S4, calculating the discrete impedance sequence measurement value of the lithium battery in the frequency domain; Step S4.1, take the nth moment and the mth moment The voltage and current data between the sudden pulse and the pulse end time are calculated, and the impedance discrete sequence of the relaxation time distribution model DRT of the lithium battery in the time domain t is calculated. ,in, The relaxation time distribution model DRT is represented in Moment The impedance under ; In the relaxation time distribution model DRT, the infinite polarization process is connected in series. Moment The impedance under ; Step S4.2: Using discrete Fourier transform Transform and write each frequency domain component after transformation into complex form, so as to obtain the impedance discrete sequence of lithium battery in frequency domain f ,in, The relaxation time distribution model DRT is represented in Frequency The impedance under , express The real part of express The imaginary part of represents the imaginary unit, ; Step S5: Fitting and solving the distribution function of the relaxation time distribution model DRT ;in, represents the time constant; Step S5.1, using formula (2) to establish the relaxation time distribution model DRT in the first Frequency The distribution function under The fitting equation is: (2) In formula (2), represents the differential symbol; Step S5.2: Use the radial basis function RBF with a Gaussian kernel function to transform the distribution function at each frequency Discretize the DRT function Converted into the sum of a series of basis functions with different weights, and the distribution function shown in formula () is obtained The fitting equations are: (3) In formula (3), and yes The real and imaginary parts of and The dimension is The real and imaginary matrices of The length of , and N = mn, M is The number of basis functions after discretization; is the weight vector of the M basis functions to be solved; Step S5.3: Solve the fitting equations using the least squares method with regularization term to obtain the optimal weight vector , and then get the DRT distribution function ; Step S6: DRT distribution function The state of charge or health state of the lithium battery is used as the input of the neural network and the output of the neural network, so as to train the neural network and obtain a state of charge prediction model or a health state prediction model of the lithium battery state parameters.

2. An electronic device, comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the lithium battery state estimation method according to claim 1, and the processor is configured to execute the program stored in the memory.

3. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the lithium battery state estimation method according to claim 1 are executed.

Images