Battery health state estimation method and system

Abstract

SOLUTION: The invention belongs to a technical field of battery health state estimation, and provides a battery health state estimation method and system. A relaxation time distribution of a battery in a current state is predicted by obtaining impedance of the battery in a plurality of specific frequency bands, and the battery health state is estimated according to the predicted relaxation time distribution. The health state of the battery under the cycle can be accurately and quickly estimated.EFFECT: According to the invention, low-cost and high-accuracy real-time accurate monitoring of the health state of the battery can be realized, and the method has great significance for improving the reliability, safety and durability of the battery.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a method and system for estimating the state of health of a battery, and more particularly to a method and system for estimating the state of health of a battery based on relaxation time distribution prediction. [Background technology]

[0002] The statements in this section are merely intended to provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In practical applications of electric vehicles and energy storage systems, it is very important for battery management systems to achieve fast and accurate monitoring of battery state. Electrochemical impedance spectra can provide more kinetic and material interface structure information than time-domain signals such as voltage and current, and can be used as a criterion for evaluating the battery's state of health, thereby improving the battery state monitoring capabilities of battery management systems.

[0004] At present, impedance spectroscopy is widely used in the fields of battery detection and research and development, and the methods for estimating the health state based on the impedance information of the battery mainly include the following types:

[0005] 1) Fitting the impedance spectrum with an equivalent circuit or mathematical model, and characterizing the component parameters in the equivalent circuit or the relevant parameters in the mathematical model to estimate the battery health state. However, the model is computationally complex and computationally intensive, making it unsuitable for online use.

[0006] 2) Data-driven methods estimate the battery health status based on the impedance characteristics at specific frequencies in the impedance spectrum. These methods have poor self-adaptability, are sensitive to noise, and are prone to large errors depending on the operating conditions.

[0007] 3) The impedance spectrum is deconvoluted to obtain the battery's relaxation time distribution, and the battery's health state is estimated based on the peak value, peak area, etc. in the relaxation time distribution curve. However, conventional methods for calculating relaxation time are very complicated and require the collection of high-density impedance data over a wide frequency range, which requires significant experimental and time costs. Summary of the Invention

[0008] To solve the above problems, the present invention proposes a method and system for estimating the state of health of a battery. The present invention predicts the relaxation time distribution of the battery in its current state by acquiring the battery impedance in multiple specific frequency bands, and accurately and quickly estimates the state of health of the battery in that cycle according to the predicted relaxation time distribution. The present invention can achieve real-time and accurate monitoring of the battery state of health at low cost and with high accuracy, which is of great significance for improving the reliability, safety, and durability of batteries.

[0009] According to some embodiments, the present invention adopts the following technical solutions.

[0010] (1) obtaining aging cycle test results of the test power battery under different working conditions, and extracting the battery impedance spectrum after each charge / discharge; (2) Calculating the relaxation time distribution curve of the battery impedance spectrum by a regularization method, and S p peaks and S V The corresponding time constants τ of the valleys l ~τ S and τ l ~τ V Based on the corresponding S in the impedance spectrum under the current working conditions p +S V calculating frequency segments; (3) The battery impedance spectrum and the corresponding relaxation time distribution curve are normalized, and the Sp +S V of the impedance of each frequency segment

number

[0011] As an alternative embodiment, step (1) specifically includes: An aging cycle test is performed on the test power battery, and the battery impedance spectrum and capacity are measured at set charge states and temperature intervals after each charge / discharge of the battery, and a database is generated in which the battery impedance spectrum and health state in different states are in one-to-one correspondence; Prior to collecting the impedance spectrum of the battery, at least the frequency range and sampling density of the impedance spectrum are determined, including but not limited to, recording the frequency, phase angle, and amplitude value of the actually measured impedance for a single impedance sampling point.

[0012] Furthermore, the validity of the acquired battery impedance spectrum is verified by KK transformation verification, and the impedance spectrum that passes the validation is selected as the finally collected impedance spectrum for calculation of the relaxation time distribution.

[0013] As an alternative embodiment, step (2) specifically includes: performing a deconvolution calculation on the impedance spectrum by a regularization method to calculate a relaxation time distribution function for the battery impedance spectrum to obtain a relaxation time distribution curve; Based on the reaction mechanism of the battery and the angle of the geometric curve, the S in the relaxation time distribution curve p peaks and S V characterizing each valley and recording the corresponding time constants on the horizontal axis of the peak and valley bottom; and calculating a corresponding frequency in the impedance spectrum based on the time constant, converting from the time scale representation back to the frequency domain representation, finding each characteristic frequency in the impedance spectrum, and if the calculated frequency is not within a test range set in the impedance spectrum, replacing it with the nearest test frequency point within the test range of the impedance spectrum.

[0014] Furthermore, the frequency segment f is determined by the experimental frequency closest to the calculated frequency and the i experimental frequencies to its left and right. l+2i By forming the parameter i, the influence of various interferences on the relaxation time prediction can be avoided, and the parameter i can be flexibly adjusted according to the requirements of prediction accuracy and test time, and each frequency segment f l+2i and the corresponding impedance segment Z l+2i Change the length of

[0015] In an alternative embodiment, step (3) specifically includes: The total number of sampling points N for each relaxation time distribution curve output by the model according to the frequency range and sampling density of the relaxation time distribution. DRT and determine the corresponding relaxation time τ DRT and normalizing the measured impedance and relaxation time data; After normalization, the impedance data and relaxation time data were dimension-reduced and S p+S V of impedance segments

number

number

number

[0016] In an alternative embodiment, step (4) specifically includes: According to the nonlinear mapping relationship between impedance segments and relaxation times, a regression prediction model (Model) was developed. DRT and A regression prediction model was selected using a training set consisting of impedance segments and relaxation time distribution curves. DRT , and identify the model parameters. During the identification, the actual resistance values ​​obtained with different relaxation times by the regularization method are

number

number

[0017] In an alternative embodiment, step (5) specifically includes: S of the relaxation time distribution function g(τ)p Integrate the peaks to obtain S p Obtaining different polarization resistances of individual cells jointly reflects the health status of the battery; S in the relaxation time distribution curve of this type of battery p According to the mathematical relationship between the polarization resistance of the peak and the state of health (SOH), the linear mathematical expression F at different temperatures and states of charge (SOC) SOH and S p Peak polarization resistance

number

[0018] In an alternative embodiment, step (6) specifically comprises: S p +S V The impedance values ​​of the frequency segments are measured and input into the trained regression prediction model, and the current complete relaxation time distribution curve (DRT) of the battery is calculated by the trained regression prediction model. pre Predict and then smooth, Predicted relaxation time distribution curve DRT pre S in p polarization resistance

number

[0019] a data acquisition module for acquiring the aging cycle test results of the power battery under test under different working conditions and extracting the battery impedance spectrum after each charge / discharge; The relaxation time distribution curve of the battery impedance spectrum is calculated by a regularization method, and S p peaks and S V The corresponding time constants τ of the valleys l ~τ S and τ l ~τ V Based on the corresponding S in the impedance spectrum under the current working conditions p +S V a frequency segment calculation module for calculating frequency segments; The battery impedance spectrum and the corresponding relaxation time distribution curve are normalized, and the S in each impedance spectrum is p +S V of the impedance of each frequency segment

number

[0020] The present invention has the following advantageous effects compared to the prior art.

[0021] The present invention selects several important segments in the relaxation time distribution curve that can best reflect (have high relevance to) the current state of the battery, develops a relaxation time distribution prediction model that can adapt to different working conditions of the battery, and uses a mathematical model to online estimate the health state of the battery based on the relationship between the polarization resistance of the battery and the health state.

[0022] The present invention proposes a method for obtaining the relaxation time distribution based on neural networks, and when calculating the relaxation time distribution at the current state of the battery, there is no need to measure the complete impedance spectrum, and no complex deconvolution or fitting calculation is required, so the calculation cost of the battery relaxation time distribution is greatly reduced.

[0023] The present invention selects appropriate features in the impedance spectrum as model input based on the reaction mechanism of the battery and the geometric structure of the relaxation time distribution curve, ensuring the predictive performance of the relaxation time prediction model while reducing the dimension of the input features. Furthermore, the length of the impedance segment to be measured can be flexibly adjusted according to the requirements of the impedance measurement time and health status estimation accuracy in practical applications, thereby expanding the applicable scenarios of relaxation time distribution technology.

[0024] The battery health state estimation method proposed by the present invention has the characteristics of being data-based and does not require complex electrochemical models. In practical application, the mathematical expression between the battery polarization resistance and the battery health state can be adjusted according to the shape of the battery relaxation time distribution curve, which improves the accuracy and practicability of battery health state estimation based on impedance information. In theory, it can be conveniently applied to various types of batteries in complex working situations.

[0025] Other advantages, objectives, and features of the invention will be set forth in part in the specification which follows, and in part will become apparent to those skilled in the art upon consideration of and examination of the following description, or may be learned from the practice of the invention. The objectives and other advantages of the invention may be realized and obtained by the following specification.

[0026] The accompanying drawings of the specification, which form a part of the present invention, are intended to provide a further understanding of the present invention, and the schematic examples of the present invention and the description thereof are intended to help interpret the present invention and are not intended to be unduly limiting of the present invention. [Brief explanation of the drawings]

[0027] [Figure 1] 1 is a flowchart of a method for estimating a battery's state of health according to an embodiment. [Figure 2] FIG. 10 is a diagram showing a distribution curve of relaxation times in step S2 according to one embodiment. [Figure 3] FIG. 1 is a diagram of frequency segments in a selected impedance spectrum according to one embodiment. [Figure 4] FIG. 10 is a diagram showing a distribution curve of relaxation times predicted based on a neural network according to an embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0028] The invention will now be further described with reference to the accompanying figures and examples.

[0029] The following detailed description is illustrative and is intended to provide further explanation of the present invention. Unless otherwise explained, all technical and scientific terms used herein have the same meaning as commonly understood by a person of ordinary skill in the art to which this invention belongs.

[0030] Furthermore, the terms used herein are merely for the purpose of describing modes for implementing the invention and are not intended to limit the exemplary embodiments of the present invention. As used herein, the singular is intended to include the plural unless the context clearly dictates otherwise. Note that, when used in this specification, the terms "comprise" and / or "comprises" indicate the presence of features, steps, operations, devices, assemblies, and / or combinations thereof.

[0031] The embodiments and features of the embodiments in the present application can be combined with each other unless they are inconsistent.

[0032] Example 1 The method for estimating the battery health is as follows: Step S1: Select a test power battery and perform an aging cycle test under different operating conditions, measure the battery impedance spectrum EIS and capacity after each charge / discharge, and verify the validity of the data; The relaxation time distribution of the impedance spectrum was calculated using the regularization method, and the S p peaks and S V The corresponding time constants τ of the valleys l ~τ S and τ l ~τ V Based on the corresponding S in the impedance spectrum under the current working conditions p +S V a step S2 of calculating frequency segments; Normalize the collected impedance spectra and the corresponding relaxation time distribution curves, and calculate S in each impedance spectrum. p +S V of the impedance of a particular frequency segment

number

number

[0033] In this embodiment, step S1 specifically includes the following steps.

[0034] In step S11, a power battery to be tested is selected and an aging cycle test is performed. After each charge / discharge of the battery, the battery impedance spectrum and capacity are measured at intervals of the set charge state ΔSOC and temperature ΔT, and a database is finally generated in which the battery impedance spectrum in different states and the state of health are in one-to-one correspondence.

[0035] In step S12, before collecting the impedance spectrum of the battery, it is necessary to determine at least, but not limited to, the frequency range and sampling density of the impedance spectrum, and also to record the frequency, phase angle, and amplitude value of the actually measured impedance for each impedance sampling point.

number

number

[0036] In step S13, the validity of the impedance spectrum needs to be verified by KK transform validation to ensure that the experimentally obtained impedance data satisfies linear stability and can be used for subsequent model training.

[0037] In this embodiment, step S2 specifically includes the following steps:

[0038] In step S21, the relaxation time distribution function g(τ) is calculated for the impedance spectrum that has passed the KK verification. The relationship between the impedance and relaxation time of the battery at different frequencies can be expressed by the following equation:

number

[0039] The calculation of g(τ) is essentially a deconvolution of known results, which can be achieved by regularization methods, resulting in the most accurate relaxation time curve. After the impedance spectrum in the frequency domain is transformed into the time domain relaxation time distribution, decoupling of the different electrochemical processes inside the battery is achieved.

[0040] In step S22, the relaxation time distribution curve diagram has several obvious peaks, and the height of the peaks reflects the intensity of the electrochemical reaction, and the position of the peaks reflects the response speed of the electrochemical reaction. Therefore, the overall shape of the relaxation time distribution curve can be basically determined by the peaks and valleys. Based on the reaction mechanism of the battery and the angle of the geometric curve, S in the diagram P peaks and S V The corresponding time constants τ on the horizontal axis of the peak and valley bottom are l ~τ S and τ l ~τ V Record the following.

[0041] In actual operation, due to different types of batteries, the number of peaks and valleys in the relaxation time distribution curve is not constant, and can only be determined by calculating the relaxation time distribution graph.

[0042] In step S23, after finding the key time constant in the relaxation time distribution, the corresponding frequency in the impedance spectrum can be calculated by the following equation:

number

[0043] By converting from the time scale representation back to the frequency domain representation, S in the impedance spectrum P +S V Individual characteristic frequency f l ~f S and f l ~f V If the calculated frequency is not within the test range set in the impedance spectrum, it is replaced with the closest frequency point within the test range of the impedance spectrum.

[0044] In step S24, when factors such as the battery's health state and cycle operation status change, there will be some drift in the time constants of different electrochemical reactions, and the experimental sampling frequency of the impedance spectrum measurement will not completely match the calculated frequency. Therefore, the frequency segment f is determined by the experimental frequency closest to the calculated frequency and the i experimental frequencies on either side of it. l+2i By forming the prediction model, it is necessary to avoid the influence of various interferences on the prediction of the relaxation time. DRT The robustness of is greatly enhanced. In practical applications, the variable i can be flexibly adjusted according to the requirements of prediction accuracy and test time, and each frequency segment f l+2i and the corresponding impedance segment Z l+2i The length can be changed.

[0045] In this embodiment, step S3 specifically includes the following steps:

[0046] In step S31, since the impedance spectrum and the relaxation time distribution curve are discrete, the S input by the model is p +S V After the ranges of the frequency segments and the corresponding impedance segments are finally determined, the total number of sampling points N of each relaxation time distribution curve output by the model is determined according to the frequency range and sampling density of the relaxation time distribution. DRT can be determined, and the corresponding relaxation time τ of each sampling point can be calculated by the formula in S23. DRT can be calculated.

[0047] In step S32, the battery is subjected to a predicted relaxation time distribution DRT because the impedance spectrum and relaxation time distribution are significantly different depending on the different states. pre To improve the accuracy of the method, the measured impedance and relaxation time data must be normalized.

[0048] In step S33, after normalization, the impedance data and relaxation time data are subjected to dimension reduction, and Sp +S V of impedance segments

number

number

number

[0049] In this embodiment, the relaxation time function g(τ) represents the relationship between the relaxation time r on the horizontal axis and the resistance value on the vertical axis, and the resistance value is

number

[0050] In this embodiment, step S4 specifically includes the following steps.

[0051] In step S41, an appropriate regression prediction model Model is calculated according to the nonlinear mapping relationship between the impedance segments and the relaxation time. DRT to predict the relaxation time curve. The type of regression prediction model selected includes, but is not limited to, a long short-term memory network (LSTM).

number

[0052] In some embodiments, the model predicts the complete relaxation time curve.

[0053] In some embodiments, the model predicts resistance values ​​for each relaxation time and forms a curve with the relaxation time function.

[0054] In step S42, a selected regression prediction model Model is calculated based on a training set consisting of impedance segments and relaxation time distribution curves. DRT The model parameters are identified, for example, by a genetic algorithm, and the Model is then trained using a gradient descent method such as self-adaptive moment estimation (Adam). DRT The parameters in the regularization method may be updated, but are not limited to: The loss function during the identification may include, but is not limited to, the root mean square error (RMSE); The actual resistance values ​​obtained by the regularization method with different relaxation times

number

number

[0055] Of course, other models may be used, and other loss functions may be used during training. Training etc. will not be described in detail here.

[0056] In this embodiment, step S5 specifically includes the following steps:

[0057] In step S51, an optimal neural network model Model DRTAfter obtaining the relaxation time distribution curve, the relaxation time distribution curve may be predicted based on the current impedance of a specific frequency segment of the target battery measured online. However, since the neural network model is nonlinear, there are sampling points in the predicted relaxation time distribution curve that cause jitter in the predicted value. For example, the predicted relaxation time distribution curve DRT pre After smoothing, the process proceeds to step S6.

[0058] In step S52, the relaxation time distribution curve calculated in steps S2 / S3 has a left peak reflecting the loss of electrical conductivity, a middle peak reflecting the loss of battery ions, and a right peak reflecting the loss of active materials of the positive and negative electrodes. Therefore, the S of the relaxation time distribution function g(τ) P Integrate the peaks to obtain S P The different polarization resistances obtained can jointly reflect the health status of the battery.

number

[0059] In step S53, S in the battery relaxation time distribution curve of that type is p According to the mathematical relationship between the peak polarization resistance and the state of health SOH, the linear mathematical expression F at different temperatures and SOC is SOH Establish.

number

[0060] In step S54, the mathematical expression of SOH may be adjusted according to the actual situation, such as the type of battery and the cycle aging environment. The types of mathematical expression include, but are not limited to, an exponential model, a sinusoidal model, and a polynomial model. The requirements for the model include the SOH and

number

number

[0061] In step S55, S p Peak polarization resistance

number

[0062] In step S56, the battery health state expression F SOH The parameters a, b, and c in the equation are fitted by the least squares method to obtain the best mathematical expression of SOH based on the polarization resistance in the battery relaxation time distribution.

[0063] In this embodiment, step S6 specifically includes the following steps.

[0064] In step S61, a specific S p +S V The impedance values ​​of specific frequency segments are measured and input to the model, and a trained regression prediction model, Model DRT Based on the current complete relaxation time distribution DRT of the battery pre Predict.

[0065] In step S62, S in the predicted relaxation time distribution curve is calculated. p polarization resistance

number

[0066] Example 2 The method for estimating the battery health state includes the following steps, as shown in FIG.

[0067] In step S1, a test power battery is selected and subjected to aging cycle tests under different working conditions, and the battery impedance spectrum EIS and capacity are measured after each charge / discharge cycle to verify the validity of the data. Specifically, the test includes the following steps:

[0068] In step S11, in one example, a test power battery is selected and an aging cycle test is performed, and the impedance spectrum and capacity of the battery are measured at intervals of 10% state of charge ΔSOC and 5°C temperature ΔT after each charge and discharge of the battery, and a database is finally generated in which the battery impedance spectrum in different states and the state of health are in one-to-one correspondence.

[0069] In step S12, the frequency range of the impedance spectrum is 2 kHz to 0.02 Hz, and the sampling density is 10 sampling points per 10 times. For each impedance sampling point, the frequency, phase angle, and amplitude value of the actually measured impedance need to be recorded.

number

number

[0070] In step S13, the validity of the impedance spectrum needs to be verified by KK transform validation to ensure that the experimentally obtained impedance data satisfies linear stability and can be used for subsequent model training.

[0071] In step S2, the relaxation time distribution of the impedance spectrum is calculated by the regularization method, and S in the relaxation time distribution curve is calculated. p peaks and S V The corresponding time constants τ of the valleys l ~τ S and τ l ~τ V Based on the corresponding S in the impedance spectrum under the current working conditions p +S V Calculating frequency segments, as shown in Figures 2 and 3, specifically includes the following steps:

[0072] In step S21, the relaxation time distribution function g(τ) is calculated for the impedance spectrum that has passed the KK verification. The relationship between the impedance and relaxation time of the battery at different frequencies can be expressed by the following equation:

number

[0073] The calculation of g(τ) is essentially a deconvolution of the known results, which can be achieved by regularization, resulting in the most accurate relaxation time curve. After the impedance spectrum in the frequency domain is transformed into the time domain relaxation time distribution, the decoupling of the different electrochemical processes inside the battery is achieved.

[0074] In step S22, as shown in FIG. 2, there are six peaks P1 to P6 and five valleys V1 to V5 from the left in the relaxation time distribution curve diagram. The height of the peaks reflects the intensity of the electrochemical reaction, and the position of the peaks reflects the response speed of the electrochemical reaction. Therefore, the overall shape of the relaxation time distribution curve can be basically determined by the peaks and valleys. Based on the reaction mechanism of the battery and the angle of the geometric curve, the six peaks and five valleys in the diagram are characterized, and the corresponding time constants τ on the horizontal axis of the peaks and valley bottoms are calculated. Pl ~τ P6 and τ Vl ~τ V5 Record the following.

[0075] In step S23, after finding the key time constant in the relaxation time distribution, the corresponding frequency in the impedance spectrum can be calculated by the following equation:

number

[0076] By converting from the time scale representation back to the frequency domain representation, the 11 characteristic frequencies f Pl ~f P6 and f Vl ~f V5 Since the corresponding frequencies of peaks P1 and P6 are not within the set test range, they are replaced with the impedance test points with the highest frequency on the left and the lowest frequency on the right of the impedance spectrum.

[0077] In step S24, when factors such as the battery's health state and cycle operation status change, there will be some drift in the time constants of different electrochemical reactions, and the experimental sampling frequency of the impedance spectrum measurement will not completely match the calculated frequency. Therefore, the frequency segment f is determined by the experimental frequency closest to the calculated frequency and one experimental frequency on either side of the calculated frequency. l+2By forming the prediction model, it is necessary to avoid the influence of various interferences on the relaxation time prediction. DRT The robustness of the

[0078] In step S3, the collected impedance spectra and the corresponding relaxation time distribution curves are normalized to obtain S in each impedance spectrum. p +S V of the impedance of a particular frequency segment

number

[0079] In step S31, since the impedance spectrum and relaxation time distribution curve are discrete, after the ranges of the 11 frequency segments input by the model and the corresponding impedance segments are finally determined, according to the frequency range and sampling density of the relaxation time distribution, the total number of sampling points of each relaxation time distribution curve output by the model is 600. The corresponding relaxation time τ of each sampling point is calculated by the formula in S23. DRT can be calculated.

[0080] In step S32, since the impedance spectrum and relaxation time distribution in different states of the battery are significantly different, the predicted relaxation time distribution DRT pre To improve the accuracy of the method, the measured impedance and relaxation time data must be normalized.

[0081] In step S33, after normalization, the impedance data and relaxation time data are subjected to dimension reduction to obtain 11 impedance segments.

number

number

[0082] In step S4, an appropriate regression prediction model is selected and offline training is performed to obtain the optimal neural network model Model DRT Specifically, the method includes the following steps:

[0083] In step S41, a long short-term memory network LSTM model is selected to predict the relaxation time curve according to the nonlinear mapping relationship between the impedance segments and the relaxation time.

number

[0084] In step S42, the LSTM model is trained using a training set consisting of impedance segments and relaxation time distributions. During training, the batch size is set to 128, the maximum number of iterations is set to 100, and the learning rate is reduced to 1 / 10 of the current learning rate every 70 iterations. The LSTM model parameters to be identified by the genetic algorithm and their identification ranges are shown in Table 1.

[0085] [Table 1]

[0086] The parameters in the LSTM model are updated using the self-adaptive moment estimation (Adam) method. The root mean square error (RMSE) is used as the loss function during classification. The actual resistance values ​​obtained with different relaxation times using the regularization method

number

number

[0087] In step S5, the characteristic peaks of the relaxation time distribution curve are integrated to obtain the polarization resistance of each peak.

number

[0088] The relaxation time distribution curve obtained in step S51 by calculation using the regularization method and prediction using the neural network model is shown in Figure 4. In the relaxation time distribution curve, the left peak reflects the loss of electrical conductivity, the middle peak reflects the loss of battery ions, and the right peak reflects the loss of active material in the positive and negative electrodes. Therefore, by integrating the six peaks of the relaxation time distribution function g(τ), six different polarization resistances can be obtained, which can jointly reflect the state of health of the battery.

number

[0089] In step S52, according to the mathematical relationship between the polarization resistance R of the six peaks in the relaxation time distribution curve of the battery of this type and the state of health SOH, a linear mathematical expression F is calculated at different temperatures and SOC. SOHEstablish.

number

[0090] In step S53, the mathematical expression of SOH may include, but is not limited to, an exponential model, a sinusoidal model, or a polynomial model depending on the type of battery and the cycle aging environment. l It is sufficient that the relationship with ~R6 can be clearly reflected while overfitting or underfitting does not occur.

[0091] In step S54, the polarization resistance R of the six peaks l ~R6 is taken as the independent variable and the battery health state SOH as the dependent variable to establish a data set of aging cycles under different working conditions for each type of battery.

[0092] In step S55, the battery health state expression F SOH By fitting the parameters in the equation (2) by the least squares method, the best-fit mathematical expression of SOH based on the polarization resistance in the battery relaxation time distribution is obtained.

[0093] In this example,

number

[0094] In step S6, the impedance of a specific frequency segment of the battery is measured online, and the trained regression prediction model Model DRT Based on the complete relaxation time distribution curve DRT pre The relationship between polarization resistance and health state F SOH The battery health status is obtained online based on the above. Specifically, the method includes the following steps:

[0095] In step S61, the impedance values ​​of 11 specific frequency segments are measured by a low-cost hardware device and input into the model, and the current complete relaxation time distribution DRT of the battery is calculated based on the trained LSTM regression prediction model. pre Predict.

[0096] Because the neural network model is nonlinear, there are sampling points that cause jitter in the predicted values ​​in the predicted relaxation time distribution. The Gaussian filter reduces the predicted relaxation time distribution curve (DRT). pre The size of the sliding window in this example is 10.

[0097] In step S62, the six polarization resistances R in the predicted relaxation time distribution curve are calculated. l Calculate ~R6 and use the expression F between polarization resistance and battery health status. SOH Based on this, the battery health status is obtained.

[0098] Example 3 The battery health estimation system a data acquisition module for acquiring the aging cycle test results of the power battery under test under different working conditions and extracting the battery impedance spectrum after each charge / discharge; The relaxation time distribution curve of the battery impedance spectrum is calculated by a regularization method, and S p peaks and S V The corresponding time constants τ of the valleys l ~τ S and τ l ~τ V Based on the corresponding S in the impedance spectrum under the current working conditions P +S V a specific frequency segment calculation module for calculating frequency segments; The battery impedance spectrum and the corresponding relaxation time distribution curve are normalized, and the S in each impedance spectrum is P +S Vof the impedance of a particular frequency segment

number

[0099] In this embodiment, the data acquisition module includes an aging cycle test device, which can be an existing device and therefore will not be described here.

[0100] Those skilled in the art will appreciate that embodiments of the present invention may be provided as a method, a system, or a computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. The present invention may also take the form of a computer program product embodied in one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0101] The present invention will be described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. Note that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams, may be implemented by computer program commands. By providing these computer program commands to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to configure an apparatus, the commands executed by the processor of the computer or other programmable data processing device may configure an apparatus for implementing the functions specified in one or more flows in the flowcharts and / or one or more blocks in the block diagrams.

[0102] These computer program commands may be stored in a computer-readable memory that can cause a computer or other programmable data processing device to operate in a particular manner, thereby configuring an article of manufacture that includes a command device, with the commands stored in the computer-readable memory implementing the functions specified in one or more flows in the flowcharts and / or one or more blocks in the block diagrams.

[0103] These computer program commands may be loaded by a computer or other programmable data processing device to cause the computer or other programmable device to perform a series of operational steps constituting a computer-implemented process, whereby the commands executed by the computer or other programmable device provide steps for implementing the functions specified in one or more flows in the flowcharts and / or one or more blocks in the block diagrams.

[0104] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent replacements, improvements, etc. made by those skilled in the art without creative labor within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. (1) obtaining aging cycle test results of a test power battery under different working conditions, and extracting the battery impedance spectrum after each charge / discharge; (2) Calculating the relaxation time distribution curve of the battery impedance spectrum by a regularization method, and S Ｐ Peaks and S Ｖ The corresponding time constants τ of the valleys ｌ ~τ Ｓ and τ ｌ ~τ Ｖ Based on this, the corresponding S in the impedance spectrum under the current working condition Ｐ +S Ｖ calculating frequency segments; (3) Normalize the battery impedance spectrum and the corresponding relaxation time distribution curve, and calculate the S Ｐ +S Ｖ of the impedance of each frequency segment [0.0000] constructing a training set using the calculated relaxation time distribution curve as an input; (4) offline training the selected regression predictive model based on the training set; (5) integrating the characteristic peaks of the calculated relaxation time distribution curve to obtain the polarization resistance of each peak, and establishing a mathematical relationship between the polarization resistance and the health state; (6) obtaining the impedance of a current battery frequency segment, predicting a relaxation time distribution curve by a trained regression prediction model, and then smoothing the predicted relaxation time distribution curve to estimate the battery's state of health based on a mathematical relationship between polarization resistance and the state of health; Specifically, step (2) includes: performing a deconvolution calculation on the impedance spectrum by a regularization method to calculate a relaxation time distribution function for the battery impedance spectrum to obtain a relaxation time distribution curve; Based on the reaction mechanism of the battery and the angle of the geometric curve, the S in the relaxation time distribution curve Ｐ Peaks and S Ｖ characterizing each valley and recording the corresponding time constants on the horizontal axis of the peak and valley bottom; Calculating corresponding frequencies in the impedance spectrum based on the time constant, converting the time scale representation back to a frequency domain representation, finding each characteristic frequency in the impedance spectrum, and if the calculated frequency is not within a test range set in the impedance spectrum, replacing it with the closest frequency point within the test range of the impedance spectrum; The frequency segment f is determined by the experimental frequency closest to the calculated frequency and the i experimental frequencies to its left and right. ｌ+2ｉ By forming the variable i, the influence of various interferences on the prediction of the relaxation time can be avoided, and the variable i can be flexibly adjusted according to the requirements of prediction accuracy and test time, and each frequency segment f ｌ+2ｉ and the corresponding impedance segment Z ｌ+2ｉ and changing the length of Specifically, step (3) includes: The total number of sampling points N for each relaxation time distribution curve output by the model according to the frequency range and sampling density of the relaxation time distribution. ＤＲＴ and determine the corresponding relaxation time τ ＤＲＴ and normalizing the measured impedance and relaxation time data; After normalization, the impedance data and relaxation time data are dimension-reduced and S Ｐ +S Ｖ of impedance segments [Equation 45] A two-dimensional data array [S Ｐ＋Ｖ (l+2i), 2] into a one-dimensional sequence [2S Ｐ＋Ｖ (l+2i),l], and then the transformed one-dimensional sequence is used as the input sequence of the model to calculate N ＤＲＴ Resistance value [Equation 46] A one-dimensional data array consisting of [Equation 47] and jointly constructing a training set for the regression prediction model using the output sequences of the model. Specifically, step (5) includes: S of the relaxation time distribution function g(τ) Ｐ The peaks are integrated to obtain S Ｐ Obtaining different polarization resistances of individual cells jointly reflects the health status of the battery; S in the battery relaxation time distribution curve Ｐ Establishing a linear mathematical expression at different temperatures and charge states according to the mathematical relationship between the polarization resistance of each peak and the state of health; S Ｐ Peak polarization resistance [Number 48] Establish a data set of aging cycles under different working conditions for each type of battery, with the independent variable being the battery health state as the dependent variable; and obtaining an optimal mathematical expression of the battery's state of health based on polarization resistance in the battery relaxation time distribution curve by fitting parameters in the expression of the battery's state of health by a least squares method; Specifically, step (6) includes: S Ｐ +S Ｖ Measure the impedance values ​​of the frequency segments, input them into the trained regression prediction model, and calculate the current complete relaxation time distribution curve (DRT) of the battery through the trained regression prediction model. ｐｒｅ Predict and then smooth, Predicted Relaxation Time Distribution Curve DRT ｐｒｅ S in Ｐ polarization resistance [Number 49] and obtaining a state of health of the battery based on an expression of the polarization resistance and the state of health of the battery. A method for estimating the state of health of a battery, comprising:

2. Specifically, the step (1) includes: An aging cycle test is performed on the test power battery, and the battery impedance spectrum and capacity are measured at set charge states and temperature intervals after each charge / discharge of the battery, and a database is generated in which the battery impedance spectrum and health state in different states are in one-to-one correspondence; 2. The method for estimating the state of health of a battery according to claim 1, further comprising: before collecting the impedance spectrum of the battery, determining a frequency range and sampling density of the impedance spectrum, and recording the frequency, phase angle, and amplitude value of the actually measured impedance for a single impedance sampling point.

3. The method for estimating the state of health of a battery according to claim 2, characterized in that the validity of the acquired battery impedance spectrum is verified by K-K transformation verification, and the impedance spectrum that passes the validity verification is selected as the finally collected impedance spectrum, and the relaxation time distribution is calculated.

4. Specifically, step (4) includes: According to the nonlinear mapping relationship between the impedance segments and the relaxation time, a regression prediction model Model ＤＲＴ and The regression prediction model Model is selected by a training set consisting of impedance segments and relaxation time distribution curves. ＤＲＴ , and identify the model parameters. During the identification, the actual resistance values ​​obtained with different relaxation times by the regularization method are [Number 50] and the resistance value predicted by the model [0.51] and calculating an error between the calculated value and the calculated value.

5. A battery health state estimation system based on the method according to any one of claims 1 to 4, comprising: a data acquisition module for acquiring the aging cycle test results of the power battery under test under different working conditions and extracting the battery impedance spectrum after each charge / discharge; The relaxation time distribution curve of the battery impedance spectrum is calculated by a regularization method, and the S ｐ Peaks and S Ｖ The corresponding time constants τ of the valleys ｌ ~τ Ｓ and τ ｌ ~τ Ｖ Based on this, the corresponding S in the impedance spectrum under the current working condition ｐ +S Ｖ a frequency segment calculation module for calculating frequency segments; The battery impedance spectrum and the corresponding relaxation time distribution curve were normalized, and the S ｐ +S Ｖ of the impedance of each frequency segment [Number 52] a training set construction module for constructing a training set using the calculated relaxation time distribution curve as an input; a model training module for performing offline training on the selected regression predictive model based on the training set; a fitting module for integrating characteristic peaks of the calculated relaxation time distribution curve to obtain the polarization resistance of each peak, and establishing a mathematical relationship between the polarization resistance and the health state; an estimation module for obtaining the impedance of a current battery frequency segment, predicting a relaxation time distribution curve by a trained regression prediction model, smoothing the predicted relaxation time distribution curve, and estimating the battery's state of health based on a mathematical relationship between the polarization resistance and the state of health; A battery health state estimation system comprising:

Images