CN118169567B

CN118169567B - An adaptive safety warning method for lithium-ion batteries based on actual operation data

Info

Publication number: CN118169567B
Application number: CN202410207980.6A
Authority: CN
Inventors: 胡晓松; 周文涛; 张凯; 李佳承; 李劲文; 庄奕; 刘弘奥; 祁清广
Original assignee: Chongqing University
Current assignee: Chongqing University
Priority date: 2024-02-26
Filing date: 2024-02-26
Publication date: 2025-03-04
Anticipated expiration: 2044-02-26
Also published as: CN118169567A

Abstract

The present invention relates to an adaptive safety early warning method for lithium-ion batteries based on actual operation data, and belongs to the field of battery diagnosis technology. The method selects the difference between the maximum and minimum voltage values and the maximum temperature as fusion feature inputs, and uses data enhancement technology to establish an enhancement matrix; selects an appropriate kernel function, maps the enhancement matrix to a high-dimensional feature space, and then calculates the enhanced kernel Mahalanobis distance; calculates the adaptive threshold according to the Chebyshev inequality, and locates the early warning time; determines the multi-level early warning value based on the median outlier value of the sliding window, and locates the abnormal single cell. The method can be applied to lithium-ion battery packs in actual operation, taking thermal runaway vehicle data as the research object, obtaining relevant parameters through normal battery data before thermal runaway occurs, and then calculating the early warning index of the vehicle at any time, and realizing timely early warning of thermal runaway by calculating multi-level adaptive thresholds.

Description

Lithium ion battery self-adaptive safety early warning method based on actual operation data

Technical Field

The invention belongs to the technical field of battery diagnosis, and relates to a lithium ion battery self-adaptive safety early warning method based on actual operation data.

Background

As an important new energy storage carrier, the lithium ion battery is an important break for solving the prominent problems of global climate change, resource environmental constraint cracking and the like, and along with the progress of technology, the global output of the lithium ion battery is continuously increased in recent years, and the industrial scale is continuously expanded. The wide application of the lithium ion battery is increasingly remarkable in the following safety problem, and news layers such as fire and explosion of the new energy automobile and the energy storage power station are endless, so that the life and property safety of people is seriously damaged, and the healthy development of the whole lithium ion battery industry is influenced. Therefore, the development of the safety early warning method of the lithium ion battery has great practical value for the whole industry.

The method for the safety early warning of the lithium ion battery mainly comprises a model-based method and a data-driven method, wherein the model-based method is required to establish an accurate lithium ion battery electric model, a thermal model or a fractional order model and the like, has high precision requirements, complex modeling process and high modeling difficulty, and diagnosis effect is influenced by the precision and the robustness of the model, the data-driven method is based on massive battery data, an artificial intelligence algorithm is developed, the signal processing method is included, the thermal runaway characteristics of BMS (battery management system) acquired data in a time domain or a frequency domain are extracted by various signal processing technologies, such as deviation, variance, frequency, entropy value, correlation coefficient and the like, and the machine learning and reinforcement learning method is also included, but the method is required to be trained by a large amount of data in advance, and the method is not interpretable.

Based on the method, the multi-dimensional signals of the battery pack are used as fusion input, the early warning index is represented by utilizing data enhancement and a kernel function, the self-adaptive threshold value is obtained in real time based on the chebyshev inequality, the early warning time is positioned, the multi-stage early warning value is determined based on the median outlier of the sliding window, and the abnormal monomer is positioned. The on-line self-adaptive safety pre-warning of the actual running lithium ion battery is realized.

Disclosure of Invention

Therefore, the invention aims to provide a lithium ion battery self-adaptive safety early warning method based on actual operation data.

In order to achieve the above purpose, the present invention provides the following technical solutions:

a self-adaptive safety pre-warning method of a lithium ion battery based on actual operation data comprises the following steps:

S1, data acquisition, namely acquiring and preprocessing the data of the battery pack with the thermal runaway accident, extracting and standardizing the information of the voltages and the temperatures of the normal segment and the abnormal segment;

s2, feature enhancement, namely taking a difference value between a maximum value and a minimum value of voltage and a maximum value of temperature as fusion features, expanding feature dimensions and establishing an enhancement matrix;

s3, calculating parameters and early warning indexes, namely calculating parameters under normal conditions based on the enhanced nuclear Markov distance, and substituting the parameters into the enhanced nuclear Markov distance of the calculated abnormal segment to serve as the early warning indexes;

S4, calculating an adaptive threshold, namely setting an early warning index sliding window while calculating an early warning index, and synchronously calculating the adaptive threshold according to the index based on chebyshev inequality;

s5, detecting fault moment, namely if the early warning index exceeds the self-adaptive threshold value, carrying out early warning, judging the fault moment according to the early warning time, and counting continuous early warning moment and duration;

S6, detecting a fault monomer, namely calculating a monomer voltage outlier based on a median outlier of the sliding window, and positioning an abnormal monomer according to the outlier degree;

and S7, grading early warning, namely combining continuous early warning time and the outlier degree of the fault monomer to prepare a multistage early warning strategy.

Further, the step S1 specifically includes the following steps:

s11, selecting an accident vehicle as a research object, acquiring vehicle normal state data and data before an accident occurs, and extracting temperature, voltage, time and alarm time data in the data;

s12, carrying out data preprocessing of removing duplication and leak, smoothing interpolation and collecting abnormal points for removing;

And S13, carrying out standardization treatment on the voltage and the temperature.

Further, the step S2 specifically includes the following steps:

s21, calculating the difference value between the maximum value and the minimum value of the voltage at the same moment, then calculating the maximum value of the temperature, taking the difference value between the maximum value of the voltage and the maximum value of the temperature as fusion characteristics, and establishing a characteristic matrix:

X=[X₁,X₂,…,X_m,...,X_D]∈R^N×D

X_m＝[x_1,m,x_2,m,...,x_k,m,...,x_N,m]

Wherein D represents a variable dimension, N represents a sampling point length, and x _k,m represents an mth dimension feature vector of a kth sampling point;

S22, introducing a process variable sample, determining the maximum time lag L of the process variable sample, wherein L is a non-negative integer, and the enhancement vector at each sampling time comprises the information of the current and the previous L sampling points so as to describe the dynamic information of the process;

s23, calculating an enhancement matrix Z with time lag L.

Further, the step S3 specifically includes the following steps:

s31: through nonlinear transformation Φ _k＝F(z_k), the enhancement matrix is mapped to a high-dimensional feature space, its dimension is denoted as h, the mapping matrix is denoted as Φ= [ Φ ₁,φ₂,...,φ_k,...,φ_N-L]∈R^h×(N-L), and the enhancement kernel mahalanobis distance of the mapping feature Φ _k at the kth sample point from the "center" of the feature space is shown:

Wherein the method comprises the steps of Representing the centered feature vector, Σ _φ representing the covariance of the mapping vector in the feature space,Representing the centering matrix to satisfyH is a symmetric idempotent matrix:

Wherein I _N-L∈R^(N-L)×(N-L) is a unit matrix of (N-L) × (N-L), and e _N-L is a unit vector of (N-L) ×1;

S32, selecting proper kernel function to calculate dot product of mapping vector to represent kernel matrix K and central kernel matrix

S33, utilizing singular value decomposition (singular value decomposition, SVD) methodPerforming full rank decomposition to calculateMoore-Penrose pseudo-inverse matrix of (E)

S34, solving a final calculation formula of the enhanced kernel Markov distance:

Order the According toThe method can obtain the following steps:

let γ=1/(N-L) HKe _N-L, then the enhancement kernel mahalanobis distance in the feature space can be calculated as follows:

Wherein the method comprises the steps of

Substituting the gamma and gamma into an enhancement matrix in a normal state, and calculating parameters gamma and gamma;

S35, substituting the enhancement matrix of the parameters gamma and the abnormal data to be detected into a calculation formula of the enhanced kernel Margaret distance, and solving an early warning index D.

Further, the step S4 specifically includes the following steps:

S41, determining the number w of sliding windows, and determining an early warning index mean mu ⁱ and a mean square error sigma ⁱ at the moment of abnormal data i to be detected;

s42, according to Chebyshev inequality, for any probability distribution, there is no more than for any real number k >0 The distribution value of (2) may differ from the average value by more than k standard deviations, and since the larger the enhanced kernel mahalanobis distance is, the more likely an abnormality exists, the upper limit of chebyshev inequality is adopted as the adaptive threshold of the algorithm, the k value is set, and the adaptive threshold is calculated in real time:

Jⁱ＝μⁱ+kσⁱ。

further, the step S5 specifically includes the following steps:

s51, comparing an early warning index of abnormal data to be detected with a self-adaptive threshold, and if the early warning index exceeds the threshold at a certain moment, locating an early warning moment;

S52, recording the starting time and the duration of continuous early warning;

Further, the step S6 specifically includes the following steps:

S61, calculating an outlier of each monomer voltage to achieve the purpose of locating abnormal monomers, and providing a median-based outlier calculation method:

Where u _ti is the voltage value normalized by the ith monomer at time t, and u _median is the median of all normalized monomer voltages at the specified window length w;

S62, setting an outlier threshold, namely according to the definition of the proposed outlier, knowing that the value range of the outlier is [0,1], wherein the closer the outlier is to 0, the more normal the monomer state is represented;

And S63, judging the outlier condition of each monomer according to the threshold value, positioning the abnormal monomer, and judging the severity of the abnormality of the monomer according to the condition that the monomer exceeds the multi-level threshold value.

The invention has the beneficial effects that:

1. The invention adopts the lithium ion battery which is actually operated as a research object, is not limited by the type and the application scene of the lithium ion battery, and has better practical significance;

2. the invention adopts fusion characteristic input and combines data enhancement technology, considers the cross correlation of multiple characteristics and the autocorrelation of single characteristic, and can more timely predict the occurrence of thermal runaway;

3. according to the invention, nonlinear change in the charge and discharge process of the lithium ion battery is fully considered, and a kernel function is introduced to better describe the early warning index;

4. the invention can early warn the lithium ion battery running in real time, and can adaptively generate the threshold value according to the state and the aging degree of the battery, thereby reducing the missing report and the false report of the early warning;

5. According to the self-adaptive multi-stage early warning value, the severity of the current abnormality of the lithium ion battery can be judged, so that maintenance personnel can carry out corresponding treatment;

6. the invention not only can early warn before the lithium ion battery is thoroughly out of control, but also can accurately locate abnormal monomers.

Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objects and other advantages of the invention may be realized and obtained by means of the instrumentalities and combinations particularly pointed out in the specification.

Drawings

For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in the following preferred detail with reference to the accompanying drawings, in which:

FIG. 1 is a technical roadmap of the invention;

FIG. 2 is a graph of the pre-processed data set of monomer voltage values and probe temperature;

FIG. 3 is a fusion feature after data enhancement;

FIG. 4 is a pre-warning indicator and adaptive threshold;

FIG. 5 is an outlier monomer maximum;

Fig. 6 is a flow chart of an implementation of the present invention.

Detailed Description

Other advantages and effects of the present invention will become apparent to those skilled in the art from the following disclosure, which describes the embodiments of the present invention with reference to specific examples. The invention may be practiced or carried out in other embodiments that depart from the specific details, and the details of the present description may be modified or varied from the spirit and scope of the present invention. It should be noted that the illustrations provided in the following embodiments merely illustrate the basic idea of the present invention by way of illustration, and the following embodiments and features in the embodiments may be combined with each other without conflict.

In which the drawings are for illustrative purposes only and are not intended to be construed as limiting the invention, and in which certain elements of the drawings may be omitted, enlarged or reduced in order to better illustrate embodiments of the invention, and not to represent actual product dimensions, it will be understood by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.

In the description of the present invention, it should be understood that, if there are terms such as "upper", "lower", "left", "right", "front", "rear", etc., the directions or positional relationships indicated are based on the directions or positional relationships shown in the drawings, only for convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred devices or elements must have a specific direction, be constructed and operated in a specific direction, so that the terms describing the positional relationships in the drawings are merely for exemplary illustration and are not to be construed as limitations of the present invention, and that the specific meanings of the terms described above may be understood by those skilled in the art according to specific circumstances.

Referring to fig. 1 to 6, the self-adaptive safety pre-warning method for the lithium ion battery based on actual operation data provided by the invention can effectively perform self-adaptive safety pre-warning on the abnormal lithium ion battery pack so as to improve the safety of the battery system, and can be divided into the following steps:

S1, data acquisition, namely acquiring and preprocessing the data of the battery pack with the thermal runaway accident, extracting and standardizing information such as voltage, temperature and the like of a normal segment and an abnormal segment;

Further, the step S1 specifically includes the following steps:

S11, selecting an accident vehicle as a research object, acquiring vehicle normal state data and data before an accident occurs, and extracting data such as temperature, voltage, time, alarm time and the like in the data;

S12, carrying out data preprocessing such as duplication elimination, leakage repairing, interpolation smoothing, abnormal point eliminating and the like on the data;

s13, carrying out standardization processing on the voltage and the temperature, wherein the voltage is taken as an example, and the standardization process is as follows:

V_t＝(v_t1,v_t2,v_t3,...,v_ti,...v_tn) (1)

U_t＝(u_t1,u_t2,u₃,...,u_tn) (4)

Wherein n is the number of battery cells in the battery pack, v _ti and U _ti respectively represent the voltage and the characteristics of the ith battery cell at the t sampling time, and U _t is the characteristic vector at the t sampling time.

Further, the step S2 specifically includes the following steps:

X=[X₁,X₂,...,X_m,...,X_D]∈R^N×D (5)

X_m＝[x_1,m,x_2,m,...,x_k,m,...,x_N,m] (6)

where D represents the variable dimension, N represents the sample point length, and x _k,m represents the kth sample point mth dimension feature vector.

S23, calculating an enhancement vector of x _k at a kth sampling point as follows:

Where x _k represents the feature vector at the kth sample point.

S24, calculating an enhancement matrix Z:

Z=[z₁,z₂,...,z_k,...,z_N-L]^T∈R^{(N-L)×(L+1)D},k>L (8)

further, the step S3 specifically includes the following steps:

Where I _N-L∈R^(N-L)×(N-L) is the identity matrix of (N-L) × (N-L), and e _N-L is the identity vector of (N-L) ×1. S32, selecting a polynomial kernel function to calculate a dot product of the mapping vector:

Where α >0, β is the polynomial degree as tuning parameter, and typically no more than 3 times to avoid overfitting.

Showing a kernel matrix K, a centered kernel matrix

K=Φ^TΦ∈R^(N-L)×(N-L) (16)

S33, utilizing singular value decomposition (singular value decomposition, SVD) methodPerforming full rank decomposition and calculatingMoore-Penrose pseudo-inverse matrix of (E)

Wherein U epsilon R ^h×h,V∈R^(N-L)×(N-L), both of which are orthogonal matrices, S epsilon R ^h×(N-L) is defined byIs comprised of singular values.

Solving a final calculation formula of the enhanced kernel mahalanobis distance:

Order the According toThe method can obtain the following steps:

Wherein the method comprises the steps of

S35, substituting the enhancement matrix of the parameters gamma and the abnormal data to be detected into a calculation formula of the enhanced kernel Margaret distance, and solving an early warning index D;

further, the step S4 specifically includes the following steps:

Jⁱ＝μⁱ+kσⁱ (25)

further, the step S5 specifically includes the following steps:

S52, recording the starting time and the duration of continuous early warning;

further, the step S6 specifically includes the following steps:

s61, calculating an outlier of each monomer voltage to achieve the purpose of locating abnormal monomers. The outlier calculation method based on the median is provided:

Where u _ti is the normalized voltage value for the ith cell at time t and u _median is the median of all normalized cell voltages at the specified window length w.

And S62, setting an outlier threshold, wherein the value range of the outlier is [0,1] according to the definition of the proposed outlier, and the closer the outlier is to 0, the more normal the monomer state is indicated. According to chebyshev inequality, calculating the self-adaptive threshold value of the maximum outlier at the same time in a sliding window mode, and setting the multi-stage early warning threshold value by adjusting different k values.

Finally, it is noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the present invention, which is intended to be covered by the claims of the present invention.

Claims

1. A lithium-ion battery adaptive safety early warning method based on actual operation data, characterized in that the method comprises the following steps:

S1: Data acquisition: obtain the battery pack data of thermal runaway accidents and perform preprocessing, extract the voltage and temperature information of normal and abnormal segments and standardize them;

S2: Feature enhancement: The difference between the maximum and minimum voltage values and the maximum temperature value are used as fusion features to expand the feature dimension and establish an enhancement matrix;

The S2 specifically includes the following steps:

S21: Calculate the difference between the maximum and minimum voltage at the same time, then calculate the maximum temperature, use the voltage maximum difference and the temperature maximum as fusion features, and establish a feature matrix:

X＝[X ₁ ,X ₂ ,…,X _m ,…,X _D ]∈R ^N×D

X _m =[x _1,m ,x _2,m ,...,x _k,m ,...,x _N,m ]

Where D represents the variable dimension, N represents the length of the sampling points, and xk _,m represents the m-th dimension feature vector of the k-th sampling point;

S22: Introduce process variable samples, determine the maximum time lag L of process variable samples, L is a non-negative integer, and the enhanced vector at each sampling time contains the information of the current and previous L sampling points to describe the dynamic information of the process;

S23: Calculate the enhancement matrix Z with a time lag of L;

S3: Calculation of parameters and early warning indicators: Calculate the parameters under normal conditions based on the enhanced kernel Mahalanobis distance, and substitute the parameters into the enhanced kernel Mahalanobis distance of the abnormal fragment as the early warning indicator;

The S3 specifically includes the following steps:

S31: Through the nonlinear transformation φ _k =F(z _k ), the enhancement matrix is mapped to a high-dimensional feature space, whose dimension is represented by h. The mapping matrix is represented as Φ=[φ ₁ ,φ ₂ ,...,φ _k ,…,φ _NL ]∈R ^h×(NL) , which represents the enhanced kernel Mahalanobis distance between the mapping feature φ _k at the kth sampling point and the “center” of the feature space:

in represents the centered eigenvector, Σ _φ represents the covariance of the mapping vector in the feature space, represents a centered matrix that satisfies H is a symmetric idempotent matrix:

Where I _NL ∈R ^(NL)×(NL) is the identity matrix of (NL)×(NL), e _NL is the unit vector of (NL)×1;

S32: Select an appropriate kernel function to calculate the dot product of the mapping vector, and express the kernel matrix K and the centered kernel matrix K:

S33: Use the singular value decomposition (SVD) method to perform full rank decomposition on Φ and calculate The Moore-Penrose pseudo-inverse matrix

S34: Solve the final calculation formula of the enhanced kernel Mahalanobis distance:

make according to You can get:

Let γ = 1/(NL)HKe _NL , then the enhanced kernel Mahalanobis distance in the feature space can be calculated as follows:

in

Substitute the normal state enhancement matrix and calculate the parameters γ and Υ;

S35: Substitute the parameters γ and Υ and the enhancement matrix of the abnormal data to be tested into the calculation formula of the enhanced kernel Mahalanobis distance to solve the early warning indicator D;

S4: Calculate the adaptive threshold: While calculating the warning indicator, set the warning indicator sliding window, and calculate the adaptive threshold synchronously according to the indicator based on the Chebyshev inequality;

S5: Fault time detection: If the warning indicator exceeds the adaptive threshold, a warning is issued, the fault time is determined according to the warning time, and the continuous warning time and duration are counted;

S6: Faulty cell detection: Based on the median outlier value of the sliding window, the cell voltage outlier value is calculated, and the abnormal cell is located according to the degree of outlier;

S7: Hierarchical warning: A multi-level warning strategy is developed based on the continuous warning time and the degree of outlier of the faulty unit.

2. The lithium-ion battery adaptive safety early warning method based on actual operation data according to claim 1, characterized in that: said S1 specifically comprises the following steps:

S11: Select the accident vehicle as the research object, obtain the normal state data of the vehicle and the data before the accident, and extract the temperature, voltage, time and alarm time data in the data;

S12: Data preprocessing including deduplication, leakage compensation, interpolation smoothing and elimination of abnormal points;

S13: Standardize the voltage and temperature.

3. The lithium-ion battery adaptive safety early warning method based on actual operation data according to claim 1, characterized in that: said S4 specifically comprises the following steps:

S41: Determine the number of sliding windows w, the mean μ ⁱ and the mean square error σ ⁱ of the warning indicators of the abnormal data to be tested at time i;

S42: According to Chebyshev's inequality, for any probability distribution, for any real number k>0, there are no more than The distribution value may differ from the mean value by more than k standard deviations. Since the larger the enhanced kernel Mahalanobis distance is, the more likely it is that an anomaly exists, the upper limit of Chebyshev’s inequality is used as the adaptive threshold of the algorithm, the k value is set, and the adaptive threshold is calculated in real time:

^Ji = ^μi + ^kσi .

4. The lithium-ion battery adaptive safety early warning method based on actual operation data according to claim 3 is characterized in that: said S5 specifically comprises the following steps:

S51: Compare the warning index of the abnormal data to be tested with the adaptive threshold. If the warning index exceeds the threshold at a certain moment, an abnormality occurs, and the warning moment is located. In actual application, when the warning continues for a certain period of time, the power supply is disconnected and corresponding rescue measures are taken;

S52: Record the start time and duration of the continuous warning.

5. The lithium-ion battery adaptive safety early warning method based on actual operation data according to claim 4, characterized in that: said S6 specifically comprises the following steps:

S61: Calculate the outlier value of each monomer voltage to locate the abnormal monomer; propose an outlier value calculation method based on the median:

Where u _ti is the standardized voltage value of the i-th monomer at time t, and u _median is the median of all standardized monomer voltages under the specified window length w;

S62: setting an outlier threshold. According to the proposed outlier definition, the value range of the outlier is [0,1]. The closer the outlier is to 0, the more normal the monomer state is. According to the Chebyshev inequality, the maximum outlier recorded at the same time is used to calculate the adaptive threshold in the form of a sliding window, and the multi-level warning threshold is set by adjusting different k values.

S63: Determine the outlier status of each monomer according to the threshold, locate the abnormal monomer, and determine the severity of the monomer abnormality according to the situation that the monomer exceeds the multi-level threshold.