CN114397577A - New energy automobile lithium battery health state assessment method based on ASTUKF-GRA-LSTM model - Google Patents
New energy automobile lithium battery health state assessment method based on ASTUKF-GRA-LSTM model Download PDFInfo
- Publication number
- CN114397577A CN114397577A CN202111500148.8A CN202111500148A CN114397577A CN 114397577 A CN114397577 A CN 114397577A CN 202111500148 A CN202111500148 A CN 202111500148A CN 114397577 A CN114397577 A CN 114397577A
- Authority
- CN
- China
- Prior art keywords
- astukf
- gra
- data
- lithium battery
- lstm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- WHXSMMKQMYFTQS-UHFFFAOYSA-N Lithium Chemical compound [Li] WHXSMMKQMYFTQS-UHFFFAOYSA-N 0.000 title claims abstract description 59
- 229910052744 lithium Inorganic materials 0.000 title claims abstract description 59
- 238000000034 method Methods 0.000 title claims abstract description 45
- 230000036541 health Effects 0.000 title claims abstract description 30
- 238000011156 evaluation Methods 0.000 claims abstract description 34
- 238000007600 charging Methods 0.000 claims abstract description 28
- 238000010219 correlation analysis Methods 0.000 claims abstract description 9
- 230000015654 memory Effects 0.000 claims abstract description 3
- 238000012549 training Methods 0.000 claims description 22
- 239000011159 matrix material Substances 0.000 claims description 21
- 238000004422 calculation algorithm Methods 0.000 claims description 16
- 238000001914 filtration Methods 0.000 claims description 14
- 230000009466 transformation Effects 0.000 claims description 13
- 238000005070 sampling Methods 0.000 claims description 12
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 238000010277 constant-current charging Methods 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000005562 fading Methods 0.000 claims description 8
- 238000005259 measurement Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 7
- 230000002829 reductive effect Effects 0.000 claims description 7
- 230000003044 adaptive effect Effects 0.000 claims description 6
- 238000010280 constant potential charging Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 238000007599 discharging Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 4
- 238000012935 Averaging Methods 0.000 claims description 3
- 241000486463 Eugraphe sigma Species 0.000 claims description 3
- 230000005284 excitation Effects 0.000 claims description 3
- 230000035772 mutation Effects 0.000 claims description 3
- 239000004576 sand Substances 0.000 claims description 3
- 230000001502 supplementing effect Effects 0.000 claims description 3
- 230000003313 weakening effect Effects 0.000 claims description 3
- 230000002441 reversible effect Effects 0.000 claims description 2
- 230000000630 rising effect Effects 0.000 claims description 2
- 238000004458 analytical method Methods 0.000 abstract description 6
- 230000015556 catabolic process Effects 0.000 abstract description 6
- 238000006731 degradation reaction Methods 0.000 abstract description 6
- 230000036961 partial effect Effects 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 230000032683 aging Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000004880 explosion Methods 0.000 description 2
- 230000007787 long-term memory Effects 0.000 description 2
- 238000000053 physical method Methods 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000002848 electrochemical method Methods 0.000 description 1
- 238000004146 energy storage Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000013210 evaluation model Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
- G01R31/367—Software therefor, e.g. for battery testing using modelling or look-up tables
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
- G01R31/378—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC] specially adapted for the type of battery or accumulator
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/36—Arrangements for testing, measuring or monitoring the electrical condition of accumulators or electric batteries, e.g. capacity or state of charge [SoC]
- G01R31/382—Arrangements for monitoring battery or accumulator variables, e.g. SoC
Abstract
The invention relates to a new energy automobile lithium battery health state evaluation method based on an ASTUKF-GRA-LSTM model, which comprises the steps of firstly obtaining a smooth IC curve, then selecting a partial area of the IC curve, and extracting battery degradation characteristics by using a gray correlation analysis method; secondly, on the basis of the input characteristic data, utilizing a long-short term memory network LSTM to carry out online estimation on the SOH of the lithium battery; finally, the method is verified to have strong SOH evaluation precision based on practical engineering experiment data. The method has higher accuracy on the SOH evaluation of the new energy automobile power, the average absolute percentage error is 0.96 percent, the root mean square error is 0.57 percent, the average evaluation time is 2.1s, and the time is far less than the time consumed by the traditional LSTM model evaluation for 8.9s and the time consumed by the GAN-CNN-LSTM model for 4.6 s. The safety analysis requirement of the direct current charging pile for accurate and rapid dynamic evaluation of the health state of the lithium battery of the new energy automobile can be effectively met. The method has high accuracy, solves the problem of low accuracy of SOH estimation of the lithium battery to a certain extent, and has certain engineering application value.
Description
Technical Field
The invention relates to a lithium battery health state evaluation method, in particular to a lithium battery health state evaluation method based on data driving, and particularly relates to a new energy automobile lithium battery health state evaluation method based on an ASTUKF-GRA-LSTM model.
Background
The lithium battery is used as an important carrier for energy storage and supply, has the characteristics Of convenience in storage, long service life and the like, is generally used as a power source Of the new energy automobile, becomes the key for long-term development Of new energy automobile strategies, and has a State Of Health (SOH) which is highly regarded by research and development personnel in the field Of new energy automobile battery safety. The lithium cell reduces the decay along with the SOH in the use, and when SOH reduced to a certain extent, can't support new energy automobile normal operating. In recent years, with the development and construction of direct current charging piles. Accurate estimation of the SOH of the lithium battery also becomes a practical requirement for safety management of the charging facility, and if the SOH estimation is inaccurate, certain economic loss is caused, and even a series of safety accidents such as charging explosion and the like are caused. Therefore, accurately mastering the health state of the current battery plays an important role in reducing the operation risk of the battery and ensuring the safe work of the direct-current charging pile. The accurate health state evaluation result of the battery not only shows the aging degree of the lithium battery, but also provides valuable reference for setting ordered charging control for charging, and is an important monitoring index for the safety management of the new energy automobile.
Due to the fact that the electrochemical mechanism of the lithium battery of the new energy automobile is complex and nonlinear, application scenes are various, and the SOH of the battery is difficult to accurately estimate through a common measurement method. Currently available battery SOH evaluation methods can be generally classified into empirical or semi-empirical model based, electrochemical/physical method based, and data-driven method based.
Electrochemical/physical based methods are also non-statistical methods, which typically utilize mathematical and physical techniques to simplify the electrochemical model to describe the health of the battery over the life of the battery. However, electrochemical models are often coupled with partial differential equations, resulting in the computational intensity required for partial differential equation calculations hindering their feasibility. Data-driven approaches are gaining more attention due to their modeless nature. These methods build battery degradation by mapping external features to battery capacity loss. In addition, some methods focus on global degradation trends and previous capabilities as input components, and thus various methods such as Support Vector Machines (SVMs), bayesian networks, auto-Regression models, and Gaussian Process Regression (GPRs) have been derived. While the data-driven approach has good non-linear characteristics and estimation accuracy, it relies on high quality data sets for effective training purposes.
Currently, some studies and documents find that the deterioration of a battery is closely related to terminal voltage during charging. The charging terminal voltage is considered to be a characteristic reflecting the residual capacity of the battery, and the comprehensive consideration of the capacity and the voltage in the charging process is a key reflecting the degradation of the battery through deep analysis. For this reason, in some research methods, Incremental Capacity (IC) is proposed as a feature for evaluating the state of health of a battery, and the IC may be calculated by differentiating a change in charge/discharge Capacity with voltage. The IC curve has higher resolution on the charging and discharging voltage plateau area, and the aging mechanism can be extracted from the peak amplitude and the position of the curve. Furthermore, IC studies have validated different angles of online SOH estimation, such as area, location, and gradient. The feasibility and accuracy of the degradation model was established using gaussian functions based on IC curves. By analyzing the capacity degradation model, estimation of the SOH of the battery is achieved. However, this method only obtains Health Features (HFs) from peaks and locations. These high frequency filters are not robust and are susceptible to noise interference. Furthermore, the proposed method cannot accommodate variable driving cycles, especially in shallow charge/discharge scenarios. In summary, the IC has good performance in achieving SOH estimation. However, there is a difficult problem with capturing the peak in the power curve, i.e. the peak is drowned out by measurement noise.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides a new energy automobile lithium battery state of health (SOH) evaluation method based on an ASTUKF-GRA-LSTM model, which is used for enhancing the charging historical data of a lithium battery through adaptive Kalman filtering, realizing accurate extraction of health performance indexes by using a grey correlation analysis method, combining with an LSTM neural network, and realizing accurate evaluation of the SOH of the lithium battery by acquiring the historical data of the new energy electric automobile based on a comprehensive energy service platform.
The technical problem to be solved by the invention is realized by adopting the following technical scheme:
a SOH (State of health) evaluation method for lithium batteries of new energy vehicles based on an ASTUKF-GRA-LSTM model comprises the following stages:
1) an adaptive strong tracking unscented Kalman filtering algorithm,
(1) the first step is to initialize, and set the initial value of the observer, namely X0，P0，Q0，R0Wherein X is0Being a matrix of state quantities, a state variable U1、U2Are respectively provided withIs the voltage across the SOC, RC network, P of the battery0To evaluate the covariance matrix, Q0As a model error matrix, R0To measure the covariance matrix and perform a priori and a posteriori estimates based on the time update and the measurement update, the state space equation of the nonlinear system is shown in equation (1), where k represents the discrete time, { W }k}、{VkIs zero mean white noise, and { W }kAnd { V }kAny elements in are uncorrelated two by two, { WkAnd { V }kThe variances of the symbols are Qk、Rk、ΓkIn order to drive the matrix for the noise,
the state space equation after nonlinear system linearization can be expressed as:
wherein the parameter matrix of the system is
UkIs excitation source, namely end current data;
(2) secondly, performing unscented transformation, selecting sampling points, namely Sigma points, near the estimation points according to rules, ensuring that the mean value and covariance of the sampling points are the same as the original state, obtaining a nonlinear function point set through nonlinear transformation, and solving the transformed mean value and covariance;
(3) and thirdly, adding an evanescence factor on the basis of a Sigma point Kalman filtering algorithm to adjust error covariance in real time so as to weaken the influence of imbalance of a battery equivalent circuit model on a measurement curve and enhance the tracking capability of state mutation, wherein the self-adaptive strong tracking unscented Kalman filtering algorithm introduces the evanescence factor in a covariance matrix:
wherein gamma isk+1The number of the cells is the fading factor,
(4) wherein the covariance of the residuals is expressed as:
wherein rho is a forgetting factor, the value range is that rho is more than 0 and less than or equal to 1, and beta is a weakening factor.
(5) The fourth step is to construct adaptive factors, a posterior suboptimal unbiased estimator proposed by Sage-Husa is adopted to update the mean value and covariance of noise, and the calculation formula is shown as the first step of formula (11), wherein dk＝(1-b)/(1-bk+1) B is a forgetting factor with a value range of 0.9-1, and G is (gamma)TΓ)-IΓT,，
2) Method for obtaining health performance index HPI by utilizing grey correlation analysis method
(1) Gray correlation analysis can quantitatively analyze the uncertain relation between the reference factors and other factors in a given system, and the capacity sequence is taken as the reference sequence X0＝{x0(k) As a comparison sequence, the sequence of HPI was marked as Xi＝{xi(k) Then the gray correlation coefficient for the ith factor can be expressed as:
wherein rho is an identification coefficient and belongs to [0,1 ]]Where p is taken to be 0.5, the grey correlation level is typically represented by averaging the values of the grey correlation coefficients, ξi(k) Comprises the following steps:
(2) selecting constant current charging stage current A1Total current A of charging in whole time period and time L of constant current charging stage1Constant voltage charging period time L2A total charging stage temperature curve T and a constant current charging stage temperature peak value T1Temperature peak T in constant voltage charging stage2Maximum slope K of the charging current curve2The capacity is estimated by 8 HPIs and used as input for the next neural network learning stage,
3) and establishing long-short term memory neural network learning based on the LSTM.
And the second step of performing the unscented transformation, the specific method comprising:
first, 2n +1 sample points, i.e. Sigma points, are selected, where n is the dimension of the state.
Then, the weights corresponding to the 2n +1 sampling points are calculated.
Finally, 2n +1 sampling points (Sigma points) and corresponding weights thereof are obtained according to the unscented transformation described by the formula (3) and the formula (4).
Moreover, the process of determining the fading factor is as follows:
the covariance of the residual sequence can be expressed as:
wherein
Definition of
The fading factor can be expressed as:
furthermore, the LSTM consists of three gates, an input gate, an output gate, and a forgetting gate.
The calculation formula of each module of the LSTM is as follows:
st＝σ[Ws(mt-1,it)+bs](formula 17)
yt＝σ[Wy(mt-1,it)+by](formula 18)
jt＝σ[Wj(mt-1,it)+bj](formula 21)
mt＝jttanh(ct) (formula 22)
In the formula: st，yt，jtWherein, the input gate, the forgetting gate and the output gate are respectively LSTM; wsAnd bs，WyAnd by，WjAnd bjAnd WcAnd bcWeights and offsets for the input gate, the forget gate, the output gate, and the cell state, respectively; m ist-1Is the output of the last moment; i.e. itInput at time t; c. CtIs a cell state;
is a candidate state; m istFor the output at time t, specifically:
(1) acquiring a historical charging and discharging data set of an electric vehicle lithium battery based on a comprehensive energy service platform as an ASTUKF-GRA-LSTM model data set, and dividing the data set into a training set and a testing set, wherein the training set accounts for 70% of data, and the testing set accounts for 30% of data;
(2) carrying out normalization processing on the training set data in the step (1),
so that all data are at [0,1 ]]To (c) to (d);
(3) taking out a sample with the capacity of m from a data set of a lithium battery
(4) From a prior distribution pz(z) taking a sample with a volume m
(5) Inputting the samples taken out from the step (3) into a generator G to obtain m generated samples
(6) Inputting the samples in the step (5) and the step (3) into a discriminator D, outputting a discrimination result, and updating network parameters of the discriminator by the discriminator according to a random gradient rising method;
(7) additionally taking out samples with the capacity of m from the prior distribution, and updating the network parameters of the generator by the generator according to a random gradient descending method;
(8) repeatedly iterating the steps (3) to (7) to enable the model to be trained stably to obtain a trained network, performing reverse normalization processing on generated data of the generator, finally generating a generated lithium battery sample with the sample capacity of m, and supplementing the generated lithium battery sample to the original data set to obtain an expanded lithium battery data set;
(9) adding the data expanded in the step (8) into an ASTUKF-GRA-LSTM model for training;
(10) training the ASTUKF-GRA-LSTM model: (a) normalizing the input data; (b) training an ASTUKF-GRA-LSTM model; (c) if the loss of the model is reduced, continuously repeating the step (b), and if the loss is not reduced continuously, starting the next step;
(11) and obtaining the trained ASTUKF-GRA-LSTM model. And (3) inputting the test set data in the step (1) into a trained ASTUKF-GRA-LSTM model, and performing inverse normalization on the output of the ASTUKF-GRA-LSTM to obtain an estimated value of the SOH.
The invention has the advantages and positive effects that:
(1) the method of the invention applies the self-adaptive strong tracking unscented Kalman filtering to carry out smoothing processing on the IC curve, overcomes the curve fault phenomenon of the statistical data of the direct current charging pile, expands the integrity of the data in the charging process, improves the tightness and reliability of the data statistics and increases the calculation precision.
(2) The method utilizes a grey correlation analysis method to extract the battery health index based on the IC curve, reduces man-made subjective interference, reserves all information quantity influencing SOH, and improves the accuracy of input data of ASTUKF-GRA-LSTM.
(3) The method provided by the invention utilizes the long-term and short-term memory neural network to efficiently train the extracted health index data, constructs an accurate evaluation model of the health state of the battery, and utilizes the real charge and discharge data of the lithium battery of the new energy automobile in the whole life cycle to verify the model.
(4) The invention provides an ASTUKF-GRA-LSTM model-based SOH online estimation model of a lithium battery of a new energy vehicle, historical data of the lithium battery is subjected to data filtering and enhancement by using an ASTUKF algorithm, and an SOH online estimation model of the lithium battery is established by combining GRA and LSTM, so that the scheme realizes quick and accurate estimation of SOH of power batteries of different models.
(5) The method adopts monitoring data of the whole life cycle of a longitudinal and transverse real new energy automobile to verify, the type of an automobile power battery is a ternary lithium battery, voltage, current, temperature and capacity data of each charge-discharge cycle of the lithium battery are extracted from the ternary lithium battery, the sample capacity is 1000, the voltage, the current and the temperature data are used as input, the first 70% of the data of the charge-discharge cycle of a data set is used as a training set, and the SOH of the lithium battery is evaluated from the rest charge-discharge cycle. LSTM model parameters: hidden nodes are 64, and the initial learning rate is 0.7. The algorithm herein is compared to the traditional LSTM model, the GAN-CNN-LSTM model. The results show that the SOH of lithium batteries decreases non-linearly with increasing cycle number. The Mean Absolute Percentage Error (MAPE) of the conventional LSTM model is 3.45%, and the Root Mean Square Error (RMSE) is 2.57%. The GAN-CNN-LSTM model corresponds to 2.1% MAPE and 1.99% RMSE. The method has high accuracy on the SOH evaluation of the new energy automobile power, the average absolute percentage error is 0.96%, the root mean square error is 0.57%, the average evaluation time is 2.1s, and the evaluation time is far less than the evaluation time of a traditional LSTM model which is 8.9s and the evaluation time of a GAN-CNN-LSTM model which is 4.6 s. The safety analysis requirement of the direct current charging pile for accurate and rapid dynamic evaluation of the health state of the lithium battery of the new energy automobile can be effectively met. The method has high accuracy, solves the problem of low accuracy of SOH estimation of the lithium battery to a certain extent, and has certain engineering application value.
Drawings
FIG. 1 is a schematic diagram of unscented Kalman filtering in the present invention;
FIG. 2 is a flow chart of the ASTUKF algorithm of the present invention;
FIG. 3 is a flow chart of SOH evaluation based on the ASTUKF-GRA-LSTM model in the present invention;
FIG. 4 is a diagram showing SOH estimation results corresponding to the models of the present invention.
Detailed Description
The present invention is further illustrated by the following specific examples, which are intended to be illustrative, not limiting and are not intended to limit the scope of the invention.
The data driving method does not need to analyze the internal mechanism of the battery, can directly establish the relation between the input characteristics and the battery capacity through historical data, and establishes the SOH (state of health) evaluation method of the lithium battery of the new energy vehicle based on the ASTUKF-GRA-LSTM model, which comprises the following stages:
1) self-adaptive strong tracking unscented Kalman filtering algorithm
(1) The first step is to initialize, and set the initial value of the observer, namely X0，P0，Q0，R0Wherein X is0Is a matrix of state quantities (state variables U)1、U2The voltage across the SOC and RC network, respectively, of the battery), P0To evaluate the covariance matrix, Q0As a model error matrix, R0To measure the covariance matrix and perform a priori and a posteriori estimates based on the time update and the measurement update, the state space equation of the nonlinear system is shown in equation (1), where k represents the discrete time, { W }k}、{VkIs zero mean white noise, and { W }kAnd { V }kAny elements in are uncorrelated two by two, { WkAnd { V }kThe variances of the symbols are Qk、Rk、ΓkThe matrix is driven for noise.
The state space equation after nonlinear system linearization can be expressed as:
wherein the parameter matrix of the system is
UkIs the excitation source, i.e., the end current data.
(2) The second step is to carry out unscented transformation, select sampling points, namely Sigma points, near the estimation points according to rules, and ensure that the mean value and covariance of the sampling points are the same as the original state, and through nonlinear transformation,
a nonlinear function point set is obtained, the mean and covariance after transformation are obtained, and the principle of the unscented transformation is shown in fig. 1.
First, 2n +1 sample points, i.e. Sigma points, are selected, where n is the dimension of the state.
Then, the weights corresponding to the 2n +1 sampling points are calculated.
Finally, 2n +1 sampling points (Sigma points) and their corresponding weights are obtained according to the unscented transformation described by equation (23) and equation (24).
(3) And thirdly, adding an evanescence factor on the basis of a Sigma point Kalman filtering algorithm to adjust the error covariance in real time so as to weaken the influence of the maladjustment of the battery equivalent circuit model on a measurement curve and enhance the capability of tracking the state mutation. The self-adaptive strong tracking unscented Kalman filtering algorithm introduces an evasive factor in a covariance matrix:
wherein gamma isk+1The determination process for the fading factor is as follows:
the covariance of the residual sequence can be expressed as:
wherein
Definition of
The fading factor can be expressed as:
(4) wherein the covariance of the residuals is expressed as:
wherein rho is a forgetting factor, the value range is that rho is more than 0 and less than or equal to 1, and beta is a weakening factor.
(5) The fourth step is to construct adaptive factors, a posterior suboptimal unbiased estimator proposed by Sage-Husa is adopted to update the mean value and covariance of noise, and the calculation formula is shown as the first step of formula (8), wherein dk＝(1-b)/(1-bk+1) B is a forgetting factor with a value range of 0.9-1, and G is (gamma)TΓ)-IΓT。
The ASTUKF algorithm flow diagram is shown in figure 2,
2) obtaining Health Performance Index (HPI) Using Gray correlation analysis
(1) Grey correlation analysis can quantify the uncertainty between a reference factor and other factors in a given system. Considering the capacity sequence as reference sequence X0＝{x0(k) As a comparison sequence, the sequence of HPI was marked as Xi＝{xi(k) Then the gray correlation coefficient for the ith factor can be expressed as:
wherein rho is an identification coefficient and belongs to [0,1 ]]Where ρ is 0.5. The gray correlation level is typically represented by averaging the values of the gray correlation coefficients, ξi(k) Comprises the following steps:
degree of gray correlation gammaiFor measuring the degree of association between the reference and the comparability sequence. Gamma rayiThe larger the correlation level. If gamma isiEqual to 1, the two sequences are identical. Therefore, we can verify and select the health performance index closely related to the capacity fade according to the grey correlation level.
(2) Although the grey correlation degree of different charge and discharge curves is different under different experimental conditions, some HPIs still have higher correlation degree obviously. In order to ensure the robustness and accuracy of HPI to different batteries and conditions, a current A in a constant-current charging stage is selected1Total current A of charging in whole time period and time L of constant current charging stage1Constant voltage charging phase time/L2A total charging stage temperature curve T and a constant current charging stage temperature peak value T1Temperature peak T in constant voltage charging stage2Maximum slope K of the charging current curve2The capacity is estimated by 8 HPIs and used as input for the next neural network learning stage.
3) Neural network learning based on long and short term memory
LSTM is a variant of the Recurrent Neural Network (RNN) that solves the problems of gradient explosion and gradient disappearance present with RNN, commonly used for time series evaluation. The LSTM consists of three gates, an input gate, an output gate, and a forgetting gate. The calculation formula of each module of the LSTM is as follows:
st＝σ[Ws(mt-1,it)+bs](formula 17)
yt＝σ[Wy(mt-1,it)+by](formula 18)
jt＝σ[Wj(mt-1,it)+bj](formula 21)
mt＝jttanh(ct) (formula 22)
In the formula: st，yt，jtWherein, the input gate, the forgetting gate and the output gate are respectively LSTM; wsAnd bs，WyAnd by，WjAnd bjAnd WcAnd bcWeights and offsets for the input gate, the forget gate, the output gate, and the cell state, respectively; m ist-1Is the output of the last moment; i.e. itInput at time t; c. CtIs a cell state;
is a candidate state; m istIs the output at time t.
(1) The method comprises the steps of obtaining a historical charging and discharging data set of the lithium battery of the electric vehicle as an ASTUKF-GRA-LSTM model data set based on a comprehensive energy service platform, and dividing the data set into a training set and a testing set, wherein the training set accounts for 70% of data, and the testing set accounts for 30% of data.
(2) Carrying out normalization processing on the training set data in the step (1),
so that all data are at [0,1 ]]In the meantime.
(3) Taking out a sample with the capacity of m from a data set of a lithium battery
(4) From a prior distribution pz(z) taking a sample with a volume m
(5) Inputting the samples taken out from the step (3) into a generator G to obtain m generated samples
(6) And (4) inputting the samples in the step (5) and the step (3) into a discriminator D, and outputting the discrimination result. The arbiter updates the arbiter network parameters according to a random gradient ascent method.
(7) And additionally taking a sample with the capacity of m from the prior distribution, and updating the network parameters of the generator by the generator according to a random gradient descent method.
(8) And (5) iterating the step (3) to the step (7) repeatedly, so that the model training is stable. And obtaining a trained network, performing inverse normalization processing on the generated data of the generator, finally generating a generated lithium battery sample with the sample capacity of m, and supplementing the generated lithium battery sample to the original data set to obtain an expanded lithium battery data set.
(9) And (4) adding the data expanded in the step (8) into an ASTUKF-GRA-LSTM model for training.
(10) Training the ASTUKF-GRA-LSTM model: (a) normalizing the input data; (b) training an ASTUKF-GRA-LSTM model; (c) if the loss of the model is reduced, the step (b) is continuously repeated, and if the loss is not reduced continuously, the next step is started.
(11) And obtaining the trained ASTUKF-GRA-LSTM model. And (3) inputting the test set data in the step (1) into a trained ASTUKF-GRA-LSTM model, and performing inverse normalization on the output of the ASTUKF-GRA-LSTM to obtain an estimated value of the SOH. The overall flow is shown in fig. 3.
The invention provides an ASTUKF-GRA-LSTM model-based SOH online estimation model of a lithium battery of a new energy vehicle, historical data of the lithium battery is subjected to data filtering and enhancement by using an ASTUKF algorithm, and an SOH online estimation model of the lithium battery is established by combining GRA and LSTM, so that the scheme realizes quick and accurate estimation of SOH of power batteries of different models.
The method adopts monitoring data of the whole life cycle of a longitudinal and transverse real new energy automobile to verify, the type of an automobile power battery is a ternary lithium battery, voltage, current, temperature and capacity data of each charge-discharge cycle of the lithium battery are extracted from the ternary lithium battery, the sample capacity is 1000, the voltage, the current and the temperature data are used as input, the first 70% of the data of the charge-discharge cycle of a data set is used as a training set, and the SOH of the lithium battery is evaluated from the rest charge-discharge cycle. LSTM model parameters: hidden nodes are 64, and the initial learning rate is 0.7. The SOH evaluation results of the algorithm are shown in FIG. 3, comparing the algorithm with the conventional LSTM model and the GAN-CNN-LSTM model.
The results show that the SOH of lithium batteries decreases non-linearly with increasing cycle number. The average absolute percent Error (MAPE) of the conventional LSTM model is 3.45%, and the Root Mean Square Error (RMSE) is 2.57%. The GAN-CNN-LSTM model corresponds to 2.1% MAPE and 1.99% RMSE. The method has high accuracy in SOH evaluation of new energy vehicles, the average absolute percentage error is 0.96%, the root mean square error is 0.57%, the average evaluation time is 2.1s, and the evaluation time is far less than the evaluation time of a traditional LSTM model which is 8.9s, and the evaluation time of a GAN-CNN-LSTM model which is 4.6 s. In conclusion, the safety analysis method and the safety analysis system can effectively meet the safety analysis requirement of the direct current charging pile on accurate, rapid and dynamic evaluation of the health state of the lithium battery of the new energy automobile. The method has high accuracy, solves the problem of low accuracy of SOH estimation of the lithium battery to a certain extent, and has certain engineering application value.
Although the embodiments of the present invention have been disclosed for illustrative purposes, those skilled in the art will appreciate that: various substitutions, changes and modifications are possible without departing from the spirit and scope of the invention and the appended claims, and therefore the scope of the invention is not limited to the embodiments disclosed.
Claims (4)
1. A SOH (State of health) evaluation method for lithium batteries of new energy vehicles based on an ASTUKF-GRA-LSTM model is characterized by comprising the following steps: the method comprises the following steps:
1) an adaptive strong tracking unscented Kalman filtering algorithm,
(1) the first step is to initialize, and set the initial value of the observer, namely X0，P0，Q0，R0Wherein X is0Being a matrix of state quantities, a state variable U1、U2The voltage across the SOC and RC network of the battery, P0To evaluate the covariance matrix, Q0As a model error matrix, R0To measure the covariance matrix and perform a priori and a posteriori estimates based on the time update and the measurement update, the state space equation of the nonlinear system is shown in equation (1), where k represents the discrete time, { W }k}、{VkIs asWhite noise of zero mean, and { WkAnd { V }kAny elements in are uncorrelated two by two, { WkAnd { V }kThe variances of the symbols are Qk、Rk、ΓkIn order to drive the matrix for the noise,
the state space equation after nonlinear system linearization can be expressed as:
wherein the parameter matrix of the system is
UkIs excitation source, namely end current data;
(2) secondly, performing unscented transformation, selecting sampling points, namely Sigma points, near the estimation points according to rules, ensuring that the mean value and covariance of the sampling points are the same as the original state, obtaining a nonlinear function point set through nonlinear transformation, and solving the transformed mean value and covariance;
(3) and thirdly, adding an evanescence factor on the basis of a Sigma point Kalman filtering algorithm to adjust error covariance in real time so as to weaken the influence of imbalance of a battery equivalent circuit model on a measurement curve and enhance the tracking capability of state mutation, wherein the self-adaptive strong tracking unscented Kalman filtering algorithm introduces the evanescence factor in a covariance matrix:
wherein gamma isk+1The number of the cells is the fading factor,
(4) wherein the covariance of the residuals is expressed as:
wherein rho is a forgetting factor, the value range is that rho is more than 0 and less than or equal to 1, and beta is a weakening factor.
(5) The fourth step is to construct adaptive factors, a posterior suboptimal unbiased estimator proposed by Sage-Husa is adopted to update the mean value and covariance of noise, and the calculation formula is shown as the first step of formula (11), wherein dk＝(1-b)/(1-bk+1) B is a forgetting factor with a value range of 0.9-1, and G is (gamma)TΓ)-IΓT,，
2) Method for obtaining health performance index HPI by utilizing grey correlation analysis method
(1) Gray correlation analysis can quantitatively analyze the uncertain relation between the reference factors and other factors in a given system, and the capacity sequence is taken as the reference sequence X0＝{x0(k) As a comparison sequence, the sequence of HPI was marked as Xi＝{xi(k) Then the gray correlation coefficient for the ith factor can be expressed as:
wherein rho is an identification coefficient and belongs to [0,1 ]]Where p is taken to be 0.5, the grey correlation level is typically represented by averaging the values of the grey correlation coefficients, ξi(k) Comprises the following steps:
(2) selecting constant current charging stage current A1Total current A of charging in whole time period and time L of constant current charging stage1Constant voltage charging period time L2A total charging stage temperature curve T and a constant current charging stage temperature peak value T1Temperature peak T in constant voltage charging stage2Maximum slope K of the charging current curve2The capacity is estimated by 8 HPIs and used as input for the next neural network learning stage,
3) and establishing long-short term memory neural network learning based on the LSTM.
2. The SOH (State of health) evaluation method of lithium batteries of new energy vehicles based on the ASTUKF-GRA-LSTM model according to claim 1, characterized in that: and the second step of performing unscented transformation, wherein the specific method comprises the following steps:
first, 2n +1 sample points, i.e. Sigma points, are selected, where n is the dimension of the state.
Then, the weights corresponding to the 2n +1 sampling points are calculated.
Finally, 2n +1 sampling points (Sigma points) and corresponding weights thereof are obtained according to the unscented transformation described by the formula (3) and the formula (4).
3. The SOH (State of health) evaluation method of lithium batteries of new energy vehicles based on the ASTUKF-GRA-LSTM model according to claim 1, characterized in that: the process of determining the fading factor is as follows:
the covariance of the residual sequence can be expressed as:
wherein
Definition of
The fading factor can be expressed as:
4. the SOH (State of health) evaluation method of lithium batteries of new energy vehicles based on the ASTUKF-GRA-LSTM model according to claim 1, characterized in that: the LSTM consists of three gates, an input gate, an output gate, and a forgetting gate.
The calculation formula of each module of the LSTM is as follows:
st＝σ[Ws(mt-1,it)+bs](formula 17)
yt＝σ[Wy(mt-1,it)+by](formula 18)
jt＝σ[Wj(mt-1,it)+bj](formula 21)
mt＝jttanh(ct) (formula 22)
In the formula: st，yt，jtWherein, the input gate, the forgetting gate and the output gate are respectively LSTM; wsAnd bs，WyAnd by，WjAnd bjAnd WcAnd bcWeights and offsets for the input gate, the forget gate, the output gate, and the cell state, respectively; m ist-1Is the output of the last moment; i.e. itInput at time t; c. CtIs a cell state;
is a candidate state; m istFor the output at time t, specifically:
(1) acquiring a historical charging and discharging data set of an electric vehicle lithium battery based on a comprehensive energy service platform as an ASTUKF-GRA-LSTM model data set, and dividing the data set into a training set and a testing set, wherein the training set accounts for 70% of data, and the testing set accounts for 30% of data;
(2) carrying out normalization processing on the training set data in the step (1),
so that all data are at [0,1 ]]To (c) to (d);
(3) taking out a sample with the capacity of m from a data set of a lithium battery
(4) From a prior distribution pz(z) taking a sample with a volume m
(5) Inputting the samples taken out from the step (3) into a generator G to obtain m generated samples
(6) Inputting the samples in the step (5) and the step (3) into a discriminator D, outputting a discrimination result, and updating network parameters of the discriminator by the discriminator according to a random gradient rising method;
(7) additionally taking out samples with the capacity of m from the prior distribution, and updating the network parameters of the generator by the generator according to a random gradient descending method;
(8) repeatedly iterating the steps (3) to (7) to enable the model to be trained stably to obtain a trained network, performing reverse normalization processing on generated data of the generator, finally generating a generated lithium battery sample with the sample capacity of m, and supplementing the generated lithium battery sample to the original data set to obtain an expanded lithium battery data set;
(9) adding the data expanded in the step (8) into an ASTUKF-GRA-LSTM model for training;
(10) training the ASTUKF-GRA-LSTM model: (a) normalizing the input data; (b) training an ASTUKF-GRA-LSTM model; (c) if the loss of the model is reduced, continuously repeating the step (b), and if the loss is not reduced continuously, starting the next step;
(11) and obtaining the trained ASTUKF-GRA-LSTM model. And (3) inputting the test set data in the step (1) into a trained ASTUKF-GRA-LSTM model, and performing inverse normalization on the output of the ASTUKF-GRA-LSTM to obtain an estimated value of the SOH.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111500148.8A CN114397577A (en) | 2021-12-09 | 2021-12-09 | New energy automobile lithium battery health state assessment method based on ASTUKF-GRA-LSTM model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111500148.8A CN114397577A (en) | 2021-12-09 | 2021-12-09 | New energy automobile lithium battery health state assessment method based on ASTUKF-GRA-LSTM model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114397577A true CN114397577A (en) | 2022-04-26 |
Family
ID=81227727
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111500148.8A Pending CN114397577A (en) | 2021-12-09 | 2021-12-09 | New energy automobile lithium battery health state assessment method based on ASTUKF-GRA-LSTM model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114397577A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115343621A (en) * | 2022-07-27 | 2022-11-15 | 山东科技大学 | Power battery health state prediction method and device based on data driving |
| CN116299005A (en) * | 2023-02-07 | 2023-06-23 | 江南大学 | Power battery health state prediction method based on AAF and deep learning |
| CN117141262A (en) * | 2023-11-01 | 2023-12-01 | 成都赛力斯科技有限公司 | Battery charging method and device and vehicle |
-
2021
- 2021-12-09 CN CN202111500148.8A patent/CN114397577A/en active Pending
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115343621A (en) * | 2022-07-27 | 2022-11-15 | 山东科技大学 | Power battery health state prediction method and device based on data driving |
| CN115343621B (en) * | 2022-07-27 | 2024-01-26 | 山东科技大学 | Method and equipment for predicting health state of power battery based on data driving |
| CN116299005A (en) * | 2023-02-07 | 2023-06-23 | 江南大学 | Power battery health state prediction method based on AAF and deep learning |
| CN116299005B (en) * | 2023-02-07 | 2023-09-05 | 江南大学 | Power battery health state prediction method based on AAF and deep learning |
| CN117141262A (en) * | 2023-11-01 | 2023-12-01 | 成都赛力斯科技有限公司 | Battery charging method and device and vehicle |
| CN117141262B (en) * | 2023-11-01 | 2024-01-23 | 成都赛力斯科技有限公司 | Battery charging method and device and vehicle |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chang et al. | Lithium-ion battery state of health estimation using the incremental capacity and wavelet neural networks with genetic algorithm | |
| CN111443294B (en) | Method and device for indirectly predicting remaining life of lithium ion battery | |
| Wei et al. | Remaining useful life prediction and state of health diagnosis for lithium-ion batteries using particle filter and support vector regression | |
| Lin et al. | State of health estimation of lithium-ion battery based on an adaptive tunable hybrid radial basis function network | |
| CN114397577A (en) | New energy automobile lithium battery health state assessment method based on ASTUKF-GRA-LSTM model | |
| CN111323719A (en) | Method and system for online determination of health state of power battery pack of electric automobile | |
| CN112765772A (en) | Power battery residual life prediction method based on data driving | |
| Li et al. | On-line estimation method of lithium-ion battery health status based on PSO-SVM | |
| Hong et al. | An iterative model of the generalized Cauchy process for predicting the remaining useful life of lithium-ion batteries | |
| CN114167284B (en) | Lithium battery RUL prediction method and equipment based on BMS big data and integrated learning | |
| CN112684363A (en) | Lithium ion battery health state estimation method based on discharge process | |
| CN114441984A (en) | Lithium battery health state estimation method | |
| Zhao et al. | State-of-health estimation with anomalous aging indicator detection of lithium-ion batteries using regression generative adversarial network | |
| Ardeshiri et al. | Gated recurrent unit least-squares generative adversarial network for battery cycle life prediction | |
| Kim et al. | State of health estimation of li-ion batteries using multi-input lstm with optimal sequence length | |
| CN113033104B (en) | Lithium battery state of charge estimation method based on graph convolution | |
| Hatherall et al. | Remaining discharge energy estimation for lithium-ion batteries using pattern recognition and power prediction | |
| CN116840724A (en) | Lithium ion battery health condition estimation and residual service life prediction method | |
| CN116736171A (en) | Lithium ion battery health state estimation method based on data driving | |
| Sun et al. | State-of-health estimation for lithium-ion battery using model-based feature optimization and deep extreme learning machine | |
| Gao et al. | Quantum assimilation-based data augmentation for state of health prediction of lithium-ion batteries with peculiar degradation paths | |
| Wang et al. | Research on SOC estimation method of hybrid electric vehicles battery based on the grey wolf optimized particle filter | |
| CN111537889B (en) | Data-driven echelon battery RUL prediction and classification method | |
| CN115598546A (en) | Combined estimation method for SOH, SOC and RUL of lithium ion battery | |
| Zhang et al. | Online state of health estimation for lithium-ion batteries based on gene expression programming |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |