CN109917292B - Lithium ion battery life prediction method based on DAUPF - Google Patents

Lithium ion battery life prediction method based on DAUPF Download PDF

Info

Publication number
CN109917292B
CN109917292B CN201910241178.8A CN201910241178A CN109917292B CN 109917292 B CN109917292 B CN 109917292B CN 201910241178 A CN201910241178 A CN 201910241178A CN 109917292 B CN109917292 B CN 109917292B
Authority
CN
China
Prior art keywords
value
new
daupf
algorithm
point set
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.)
Expired - Fee Related
Application number
CN201910241178.8A
Other languages
Chinese (zh)
Other versions
CN109917292A (en
Inventor
袁慧梅
朱骏
谭天雄
吴立锋
宋宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Zhonglian Technology Service Co ltd
Guangxi Jubang Energy Co ltd
Original Assignee
Capital Normal University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Capital Normal University filed Critical Capital Normal University
Priority to CN201910241178.8A priority Critical patent/CN109917292B/en
Publication of CN109917292A publication Critical patent/CN109917292A/en
Application granted granted Critical
Publication of CN109917292B publication Critical patent/CN109917292B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Tests Of Electric Status Of Batteries (AREA)

Abstract

The invention relates to a lithium ion battery life prediction method based on DAUPF, which comprises the steps of firstly, sampling a part, adding a double self-adaptive factor on the basis of a UKF algorithm, taking a state value and covariance obtained after one-step prediction of a Sigma point set as a guide, carrying out UT (inverse transform) conversion again to obtain a new Sigma point set, and bringing the new Sigma point set into an observation equation to obtain a new observed quantity so as to obtain a first cycle sample mean value and covariance; and after the UKF algorithm improvement part finishes one cycle, updating an adaptive factor, and then performing the next UKF algorithm improvement cycle. Entering a PF process after sampling is finished, updating another adaptive factor after a primary output predicted value is obtained, and finishing a DAUPF process; and finally predicting test data. The invention improves the UPF algorithm sampling part, the addition of double adaptive factors enables the algorithm to have stronger robustness, the two steps of UT conversion enable the adaptive factors to be better integrated into the algorithm, and the algorithm prediction effect is more accurate.

Description

Lithium ion battery life prediction method based on DAUPF
Technical Field
The invention relates to a lithium ion battery service life prediction method based on DAUPF (double-adaptive sampling electrodeless Kalman particle filter algorithm), belonging to the technical field of lithium battery health management.
Background
Lithium ion batteries have been successfully used in many consumer electronics products (e.g., mobile phones, notebook computers, and electric vehicles), and have been gradually expanded to the fields of military communication, navigation, aviation, aerospace, and the like. The safety of lithium ion batteries is receiving more and more attention. The battery life is defined as the number of cycles or age of charge and discharge. The chemicals in the battery gradually degrade as the operating time of the battery increases, and battery failure can have serious consequences. The U.S. California fire department states that a Tesla Model S vehicle auto-ignites in a parking lot and reignites in a trailer yard after a few hours, without collision and other operations of the vehicle during the two auto-ignitions. It is very important to accurately predict the service life of the lithium ion battery. State of health estimation (SOH) and remaining life prediction are key to battery health management, which can ensure safe use of lithium ion batteries.
Currently, there are two types of prediction methods for lithium batteries. One is a nonparametric model method, and the other is a parametric model method.
Examples of the nonparametric method include a neural network method and a machine learning method. Wu et al estimate the Remaining Useful Life (RUL) of a lithium ion battery using a Feed Forward Neural Network (FFNN) and Monte Carlo (IS) method. Zhang et al used a neural network based on long-short term memory (LSTM) to predict the remaining useful life of the battery. Examples of the machine learning method include support vector classification (SVM), Support Vector Regression (SVR), and Relevance Vector Machine (RVM) method. Tobar et al apply a method of improved nuclear adaptive filtering to the prediction of electric bicycle battery voltage.
Parametric models are most commonly used in various filtering algorithms. The Particle Filter (PF) method is an approximate Bayesian filter algorithm based on Monte Carlo simulation. The core idea is to use discrete random sampling points to approximate the probability density function of the system random variable. The seedling and the like predict the residual service life of the battery by a particle filtering method, and the obtained particle filtering can well predict the residual service life of the lithium ion battery. Saha et al establish a battery system framework to predict the remaining useful life of the battery at different discharge rates via PF. The particle filtering method is applicable to any non-linear non-gaussian environment, but depends on the selected reference distribution and state a posteriori estimation.
Extended Kalman (EKF) and Unscented Kalman (UKF) are modified Kalman Filter (KF) algorithms. The EKF has the advantages of weak nonlinearity and good prediction effect in a less-noise environment. The dong et al extracted an Adaptive Extended Kalman (AEKF) algorithm based on the recursive least square method, and found that the AEKF suppressed noise well. Ramadan et al analyze and compare various EKF algorithms to obtain a parameter model that is required to predict the state of charge (SOC) of a battery, and the quality of the EKF algorithms is closely related to the accuracy of the model. The UKF has the advantages of no loss on the model and relatively high calculation precision. Zheng proposes an integrated UKF method to predict the RUL of the battery, and estimates the parameters of the battery by using future residual errors, so that the short-term capacity of the battery can be accurately predicted, but the prediction precision cannot be further improved because the UKF cannot adjust the model parameters.
The particle filtering and the Kalman filtering have advantages and disadvantages respectively, and the advantages of the two algorithms can make up for the mutual deficiency, so that the extended Kalman particle filtering method and the unscented Kalman particle filtering method are provided. Seedling et al successfully predicted the RUL of the cell by using the infinite kalman particle filter (UPF) algorithm. But the algorithm is too dependent on the number of particles, the size of the data set, and the quality of the historical data. Zhang et al can maintain particle diversity and predict the remaining life of a lithium ion battery using a UPF algorithm based on Markov-Monte Carlo. Chen et al used a second order gaussian model and UPF to predict battery life.
The UPF algorithm is used for guiding particle sampling by the UKF algorithm in a sampling stage. And performing PF algorithm step after sampling, calculating weight, and performing normalization treatment. And judging whether resampling is carried out or not, and copying and eliminating the particle set. And calculating the mean value of the particle set to obtain an estimated output value. And analyzing the data after the iteration is finished. The structure is shown in fig. 1.
Due to introduction of UKF algorithm to guide sampling, the UPF algorithm is easily influenced by constraint of Gaussian noise and reference distribution; in addition, after the conventional ut (unknown transform) transform is transformed, the state value is updated, a certain error exists between the sigma distribution of the updated state value and the sigma distribution before updating, and if the sigma distribution used before updating is used to calculate parameters such as the observation prediction value, a certain influence is also exerted on the prediction result.
Disclosure of Invention
The invention aims to provide a lithium ion battery service life prediction method based on DAUPF (digital up-conversion), aiming at solving the problem that a UPF algorithm is easily influenced by noise and reference distribution, and firstly, an adaptive factor is added into the DAUPF algorithm. Considering that the traditional UPF algorithm is the combination of UKF and PF algorithm, adaptive factors are added in the sampling stage and the prediction stage respectively, and the adaptive factors can adjust parameter distribution, thereby making up the defects of the two algorithms. Secondly, the probability density is given by using a UKF algorithm in a sampling stage, the UKF can carry out UT conversion during sampling, and sigma distribution given by the UT conversion is inaccurate due to the fact that a new self-adaptive factor needs to be added after the UT conversion in the first step. In addition, the state value is updated after the conventional UT conversion, a certain error exists between the sigma distribution of the updated state value and the sigma distribution before updating, and if parameters such as an observation predicted value and the like are calculated by using the sigma distribution proposed before updating, a certain influence is generated on a prediction result. Based on the two reasons, the DAUPF algorithm carries out UT conversion once again after the state value is updated to obtain a new sigma point set, and then the parameters such as the observation predicted value and the like are calculated. The validity of the DAUPF algorithm is verified by using lithium battery experimental data of an advanced life cycle engineering center of Maryland university, and is compared with extended Kalman filtering, unscented Kalman filtering, particle filtering, extended Kalman particle filtering and unscented Kalman particle filtering.
The invention discloses a lithium ion battery service life prediction method based on DAUPF, which adopts the following technical scheme for solving the problems: firstly, a sampling part adds a double self-adaptive factor on the basis of a UKF algorithm, then guides a Sigma point set to predict in one step to obtain a state value and a covariance, and then carries out UT conversion once to obtain a new Sigma point set which is brought into an observation equation to obtain a new observed quantity, thereby obtaining a first-cycle sample mean value and a first-cycle covariance; and after the UKF algorithm improvement part completes one cycle, updating one adaptive factor in the double adaptive factors, and then performing the next UKF algorithm improvement cycle. Entering a PF process after sampling is finished, updating another adaptive factor after a primary output predicted value is obtained, and finishing a DAUPF process; finally, test data is predicted.
The invention relates to a lithium ion battery service life prediction method based on DAUPF, which specifically comprises the following steps:
step1, initializing parameters;
step2, entering into an improved UKF to guide particle distribution;
step3. calculating Sigma point set for the first time through UT transformation to obtain
Figure GDA0002742978280000031
Step4, adding a double adaptive factor to obtain
Figure GDA0002742978280000032
Step5. calculating mean and covariance from the Sigma point set obtained at step3
Figure GDA0002742978280000033
Step6, using the mean value and covariance obtained from step5, and performing UT transformation again to obtain a new Sigma point set
Figure GDA0002742978280000034
Step7. obtaining the observation prediction value by predicting the new Sigma point set obtained by step6
Figure GDA0002742978280000035
Obtaining new observation prediction mean value by observation prediction value and state prediction value through non-trace transformation calculation
Figure GDA0002742978280000036
Mean and covariance
Figure GDA0002742978280000037
Step8, calculating Kalman gain, variance and state updating;
step9. update the first adaptive factor
Figure GDA0002742978280000038
A value;
and step10, judging whether sampling is finished or not. If the weight normalization is finished, performing the next weight normalization processing, otherwise entering step 2;
step11, calculating the weight by using the mean value and the variance obtained by the step1-9 sampling part through normalization processing to obtain a normalized weight;
step12, resampling particles; updating data, updating state, updating variance and taking the mean value as final estimation.
Obtaining a predicted value, and updating the beta value of the adaptive factor;
step14. judge whether the iteration is finished. If so, evaluating the algorithm, otherwise, entering step 2;
step15. evaluation algorithm.
Wherein, the parameters initialized by Step1 include: initialized state value
Figure GDA0002742978280000041
Wherein the content of the first and second substances,
Figure GDA0002742978280000042
in order to observe the initial state values of the equation,
Figure GDA0002742978280000043
to initialize the covariance matrix.
Step2 is a starting sampling phase, the whole sampling phase circulates for N times, and the sampling phase is as follows: step2-Step 9.
Wherein Step3 specifically comprises the following steps: calculating a Sigma Point set of 2n +1 sample points
Figure GDA00027429782800000412
Wherein, the point set
Figure GDA00027429782800000413
From point Xk-1And
Figure GDA0002742978280000044
the composition of the components, wherein,
Figure GDA00027429782800000414
as a scaling function.
Wherein Step4 specifically comprises the following steps:
Figure GDA0002742978280000045
two adaptive factors are added, and the initial values of the two adaptive factors are 1. The first adaptive factor is
Figure GDA0002742978280000046
Wherein Z isk-1Is the observed value of the previous sampling point,
Figure GDA00027429782800000415
the mean value was predicted for the observation obtained at Step7 in the previous cycle. The second adaptive factor is
Figure GDA00027429782800000416
Wherein Z isk-1For true value of the system, Zupfk-1And (4) obtaining a predicted value after the Step13 DAUPF algorithm finishes the previous cycle.
Wherein Step5 is a one-Step prediction of sampling points, and the average value is
Figure GDA00027429782800000417
By
Figure GDA00027429782800000418
And calculating to obtain the result, wherein,
Figure GDA00027429782800000419
the Sigma point set obtained in Step3 is substituted into the nonlinear transformation function. Covariance
Figure GDA0002742978280000047
Figure GDA0002742978280000048
Wherein
Figure GDA0002742978280000049
Wherein λ ═ α2(n+κ)-n；α＝1；ρ＝0；κ＝2。
Wherein Step6 is the second UT transformation, generates a new Sigma point set,
Figure GDA00027429782800000410
wherein, the point set
Figure GDA00027429782800000420
By point
Figure GDA00027429782800000421
And
Figure GDA00027429782800000411
composition of, wherein
Figure GDA00027429782800000422
The mean value obtained in Step5.
Wherein Step7 observes the predicted value
Figure GDA00027429782800000423
New Sigma Point set from Step6
Figure GDA00027429782800000424
And substituting the state equation function to obtain the target. New observed predicted mean
Figure GDA00027429782800000425
Prediction of values from observations
Figure GDA00027429782800000426
And obtaining the weight. New mean value
Figure GDA00027429782800000427
And weighting the new observation prediction mean value and the observation prediction value. New covariance
Figure GDA00027429782800000428
New Sigma Point set from Step6
Figure GDA00027429782800000429
Mean values obtained in Stap5
Figure GDA00027429782800000430
And weighting the new observation prediction mean value and the observation prediction value.
Wherein Step8 calculates Kalman gain Kk，
Figure GDA00027429782800000431
Is the new mean value in Step7
Figure GDA00027429782800000432
With new covariance
Figure GDA00027429782800000433
Product of inverse matrices. Updated system covariance
Figure GDA00027429782800000434
By Kalman gain KkWith the new mean value
Figure GDA00027429782800000435
And (4) calculating. Updated state
Figure GDA00027429782800000436
From the new Sigma point set in step6
Figure GDA00027429782800000437
Kalman gain KkObserved value Z of current sampling pointkWith new observed predicted mean
Figure GDA0002742978280000051
The difference of (a) is calculated.
Wherein, Step9 updates the first adaptive factor, and the specific steps are the same as Step4.
Wherein, the Step13 obtains a predicted value, the predicted value ZupfkAnd substituting the normalized weight value into the state equation function to obtain the normalized weight value, and updating a second self-adaptive factor.
Step14 is a judging Step for judging whether the algorithm is completed.
The invention relates to a lithium ion battery service life prediction method based on DAUPF, which has the advantages and effects that: the sampling part of the UPF algorithm is improved, the algorithm has stronger robustness due to the addition of the double adaptive factors, and the adaptive factors can be better integrated into the algorithm due to the two-step UT conversion, so that the algorithm prediction effect is more accurate.
Drawings
Fig. 1 shows a flow chart of the UPF algorithm.
FIG. 2 is a flow chart of the method of the present invention.
Fig. 3 is a graph showing the capacity change of the data A3, a5, A8 and a12 of lithium ion batteries of group 4 university of maryland.
Fig. 4a is a graph comparing the true value of a3 battery data with the results of four algorithms.
Fig. 4b shows the absolute error of the true a3 battery data versus the four algorithms.
Fig. 5a to 5d are graphs showing the error probability density of the true value of the a3 battery data and the four algorithms.
Fig. 6a shows an AME diagram of 101 cycles of a3 battery data.
Fig. 6b shows the RMSE graph for 101 cycles of a3 battery data.
Fig. 6c shows the AME of A8 battery data for 101 cycles.
Fig. 6d shows the RMSE graph for 101 cycles of A8 battery data.
FIG. 7a shows the AME values for each algorithm for different battery data.
FIG. 7b shows the RMSE values for each algorithm for different battery data.
Detailed Description
The technical solution of the present invention is further described below with reference to the accompanying drawings and examples.
As shown in fig. 2, the method for predicting the service life of a lithium ion battery based on DAUPF of the present invention specifically includes the following steps:
step one, initializing parameters, comprising: initialized state value
Figure GDA0002742978280000052
Wherein the content of the first and second substances,
Figure GDA0002742978280000053
in order to observe the initial state values of the equation,
Figure GDA0002742978280000054
to initialize the covariance matrix.
And step two, entering an improved UKF to guide the particle distribution. The method comprises the following steps of starting a sampling stage, circulating the whole sampling stage for N times, and carrying out the whole sampling stage: from step two to step nine.
Step three, calculating a Sigma point set for the first time through UT conversion to obtain
Figure GDA00027429782800000612
The method specifically comprises the following steps: calculating a Sigma Point set of 2n +1 sample points
Figure GDA00027429782800000613
Wherein, the point set
Figure GDA0002742978280000061
From point Xk-1And
Figure GDA0002742978280000062
the composition of the components, wherein,
Figure GDA00027429782800000614
as a scaling function.
Step four, adding a double self-adaptive factor to obtain
Figure GDA00027429782800000615
The method specifically comprises the following steps: two adaptive factors are added, and the initial values of the two adaptive factors are 1. The first adaptive factor is
Figure GDA0002742978280000063
Wherein Z isk-1Is the observed value of the previous sampling point,
Figure GDA00027429782800000616
the mean is predicted for the observations made at step seven in the previous cycle. The second adaptive factor is
Figure GDA00027429782800000617
Wherein Z isk-1For true value of the system, Zupfk-1And a predicted value obtained after the thirteen DAUPF algorithm completes the cycle of the previous period is obtained.
Step five, calculating the mean value and the covariance
Figure GDA0002742978280000064
Step five is one-step prediction of sampling points, and the average value is
Figure GDA00027429782800000618
By
Figure GDA00027429782800000619
And calculating to obtain the result, wherein,
Figure GDA00027429782800000620
substituting the Sigma point set obtained in the third step into a nonlinear transformation function to obtain the Sigma point set; covariance
Figure GDA0002742978280000065
Wherein
Figure GDA00027429782800000621
Figure GDA0002742978280000066
Wherein λ ═ α2(n+κ)-n；α＝1；ρ＝0；κ＝2；
And sixthly, performing UT conversion again to obtain a new Sigma point set. By a second UT transformation, a new Sigma point set is generated
Figure GDA00027429782800000622
Wherein, the point set
Figure GDA0002742978280000067
By point
Figure GDA00027429782800000624
And
Figure GDA00027429782800000623
Figure GDA0002742978280000068
composition of, wherein
Figure GDA00027429782800000625
And 5, obtaining the average value in the step five.
Seventhly, predicting to obtain an observation predicted value; calculating by using the observation predicted value and the state predicted value through unscented transformation to obtain a new observation predicted mean value, a new average value and a new covariance; the method comprises the following specific steps:
observation of predicted values
Figure GDA0002742978280000069
New observed predicted mean
Figure GDA00027429782800000610
The new mean is weighted from the new observed predicted mean and the observed predicted value, i.e.
Figure GDA00027429782800000611
New covariance from the New Sigma Point set obtained in step six
Figure GDA00027429782800000626
Mean value obtained in step five
Figure GDA00027429782800000627
The new observed mean value is weighted with the observed predicted value, i.e.
Figure GDA0002742978280000071
Step eight: and calculating Kalman gain, variance and state update.
Kalman gain
Figure GDA0002742978280000074
Updated system covariance by Kalman gain KkWith the new mean value
Figure GDA0002742978280000077
Is calculated to obtain
Figure GDA0002742978280000075
Figure GDA0002742978280000076
The updated state is represented by the new Sigma point set in step six
Figure GDA0002742978280000078
Kalman gain KkObserved value of current sampling point and new observation prediction mean value
Figure GDA0002742978280000079
Is calculated as a difference of (i) i
Figure GDA0002742978280000072
Nine, updating the first adaptive factor
Figure GDA00027429782800000710
The value is obtained.
Figure GDA0002742978280000073
Wherein Z isk-1Is the observed value of the previous sampling point,
Figure GDA00027429782800000711
the mean is predicted for the observations made at step seven in the previous cycle.
And step ten, judging whether sampling is finished or not. If the weight normalization is finished, performing the next weight normalization processing, otherwise, returning to the step two;
step eleven, calculating the weight by using the mean value and the variance obtained by the sampling part completed in the step one to the step nine, and obtaining a normalized weight by normalization processing;
and step twelve, resampling particles. Updating data, updating state, updating variance and taking the mean value as final estimation;
thirteen, obtaining a predicted value and updating the adaptive factor betaThe value is obtained. Prediction value
Figure GDA00027429782800000712
Substituting the normalized weight value into the state equation function to obtain the second adaptive factor
Figure GDA00027429782800000713
Wherein Z iskFor true value of the system, ZupfkAnd a predicted value is obtained after the thirteen DAUPF algorithm completes a cycle.
And step fourteen, judging whether iteration is finished. If the algorithm is finished, evaluating the algorithm, otherwise, returning to the step two; wherein, the second step to the third step are a periodic cycle of the DAUPF algorithm, and the DAUPF algorithm is set by requirements for several times.
And fifteen, evaluating an algorithm.
The specific embodiment is as follows:
the experiment uses matlab to simulate, based on the experimental data of lithium batteries of the advanced life cycle engineering center of the university of maryland, No.03,05,08 and 12 are selected as experimental data, the experimental data of 4 groups of lithium ion batteries is shown in figure 3, and the experiment uses the same type of batteries with different capacity degradation rates and the same brand of batteries, and is carried out under the same working condition. The charge and discharge test method of the lithium battery comprises the following steps: the ArbinBT2000 battery test system is used for carrying out charge and discharge tests at room temperature, and when the charge or discharge voltage reaches the cut-off voltage specified by a manufacturer, a charge or discharge process is completed. The rated capacity of the battery is 0.9Ah, and the discharge current is 0.4 Ah.
The lower initial values a, b, c, d are the values obtained after fitting No.03,05,08, 12. The process noise and process noise variance are set to 0.0001 and 0.001, respectively. The experimental observation model uses a capacity attenuation model Zk＝a*exp(b*k)+c*exp(d*k)。
1. Setting the initial value a to-0.0000083499; 0.055237; c is 0.90097; substitution of d-0.00088543
Figure GDA00027429782800000811
To obtain
Figure GDA0002742978280000081
Figure GDA0002742978280000082
Is provided with Z0＝0.9208。
2. The sampling phase is cycled for N times
3.
Figure GDA0002742978280000083
Figure GDA0002742978280000084
Figure GDA0002742978280000085
Figure GDA0002742978280000086
4.
Figure GDA0002742978280000087
5. A second UT transformation is performed and,
Figure GDA0002742978280000088
Figure GDA0002742978280000089
Figure GDA00027429782800000810
Wj (m)＝[0.3333 0.0833 0.0833 0.0833 0.0833 0.0833 0.0833 0.0833 0.0833]
Wj (c)＝[0.3333 0.0833 0.0833 0.0833 0.0833 0.0833 0.0833 0.0833 0.0833]
Figure GDA0002742978280000091
Figure GDA0002742978280000092
6.
Figure GDA0002742978280000093
Figure GDA0002742978280000094
Figure GDA0002742978280000095
Figure GDA0002742978280000096
7.
Figure GDA0002742978280000097
Figure GDA0002742978280000098
R＝0.0001
Figure GDA0002742978280000101
Figure GDA0002742978280000102
8.
Figure GDA0002742978280000103
Figure GDA0002742978280000104
Figure GDA0002742978280000105
9. updating adaptive factor 1
Figure GDA0002742978280000106
10. Completing the sampling for 1 time, and repeating the steps 2-10 for the subsequent sampling. The above is data of only one cycle.
11. And sampling the mean value and the variance obtained by the part, and carrying out normalization processing to calculate the weight so as to obtain a normalized weight.
12.
Figure GDA0002742978280000107
Updating the adaptive factor 2
Figure GDA0002742978280000108
13. The above values are prediction data Zupf obtained by performing a DAUPF cycle, and the program needs to be continuously operated to predict more data.
The validity of the DAUPF algorithm is verified by using lithium battery experimental data of the advanced life cycle engineering center of the university of Maryland, and compared with extended Kalman filtering, unscented Kalman filtering, particle filtering, extended Kalman particle filtering and unscented Kalman particle filtering.
In order to illustrate the accuracy of the prediction effect of the DAUPF, the experiment is respectively compared with UKF, PF and UPF, and the initial parameters and error values of the four algorithms are consistent with the initial parameter values and error values of the DAUPF.
Fig. 4 and 5 correspond to battery data a3, fig. 4a and 4b show the prediction results and absolute errors, respectively, and fig. 5a to 5d are error probability density diagrams.
In the figure, the black curve represents the output true value; circles represent the predicted outcome of the PF; the square grid is a prediction result of the UKF algorithm; the diamond is a UPF algorithm result; the cross sign is the DAUPF prediction result; the horizontal line is the battery capacity failure threshold. As can be seen from fig. 4a, the prediction effect is significantly improved with the continuous improvement of the filtering method, wherein the line represented by the DAUPF algorithm is closer to the line represented by the true value. From the absolute error fig. 4b it can be seen that the absolute error of the DAUPF is minimal, the PF is least effective, and the predicted effect is worse the further up to the point of failure. It can be seen from fig. 5a to 5d that the DAUPF algorithm of the present invention is most stable and has the strongest robustness.
The experiment was completed for 101 cycles, and the resulting MAE and RMSE after each cycle were recorded and plotted as line graphs as shown in FIGS. 6 a-6 d. The closer the RMSE and MAE values are to 0, the more accurate the prediction method is. It can be seen from the figure that, for different data sets A3 and A8, different data volumes, different process noise and observation noise, the MAE value and RMSE value of the predicted value of the DAUPF algorithm are the smallest compared with other algorithms, and are twice smaller than the UPF error with the best prediction effect, and the stability of the algorithm is also stronger than that of other algorithms (from the horizontal axis, it can be seen that the DAUPF algorithm can reduce the influence of observation noise and process noise of various sizes). Therefore, the DAUPF algorithm of the present invention predicts better performance than several other algorithms.
As can be seen from fig. 7a and 7b, the PF algorithm is not ideal for predicting the a3 data set, i.e. the particle filter has a poor prediction effect on the data set with fewer data points. The UKF algorithm has a good prediction effect on data with large data fluctuation. The UPF algorithm has good prediction effect on different data sets, but the data sets with few data points have certain influence on the prediction accuracy of the UPF algorithm. The prediction effect of the DAUPF algorithm under different data sets is better than that of other algorithms, the error is minimum, the DAUPF algorithm is most stable, and the robustness is stronger.

Claims (9)

1. A lithium ion battery service life prediction method based on DAUPF is a double-adaptive sampling electrodeless Kalman particle filter algorithm and is characterized in that: the method specifically comprises the following steps:
step1, initializing parameters;
step2, entering into an improved UKF to guide particle distribution;
step3. calculating Sigma point set for the first time through UT transformation to obtain
Figure FDA0002742978270000011
Step4, adding a double adaptive factor to obtain
Figure FDA0002742978270000012
Step5. calculating mean and covariance from the Sigma point set obtained at step3
Figure FDA0002742978270000013
Step6, using the mean value and covariance obtained from step5, and performing UT transformation again to obtain a new Sigma point set
Figure FDA0002742978270000014
Step7. obtaining the observation prediction value by predicting the new Sigma point set obtained by step6
Figure FDA0002742978270000015
Obtaining new observation prediction mean value by observation prediction value and state prediction value through non-trace transformation calculation
Figure FDA0002742978270000016
Mean and covariance
Figure FDA0002742978270000017
Step8, calculating Kalman gain, variance and state updating;
step9. update the first adaptive factor
Figure FDA0002742978270000018
A value;
step10, judging whether sampling is finished or not; if the weight normalization is finished, performing the next weight normalization processing, otherwise entering step 2;
step11, calculating the weight by using the mean value and the variance obtained by the step1-9 sampling part through normalization processing to obtain a normalized weight;
step12, resampling particles; updating data, state updating, variance updating, mean as final estimate
Obtaining a predicted value, and updating the beta value of the adaptive factor;
step14, judging whether iteration is finished or not; if so, evaluating the algorithm, otherwise, entering step 2;
step15. evaluation algorithm.
2. The method of claim 1, wherein the method for predicting the lifetime of a lithium ion battery based on DAUPF comprises: the parameters initialized by Step1 include: initialized state value
Figure FDA0002742978270000019
Wherein the content of the first and second substances,
Figure FDA00027429782700000110
in order to observe the initial state values of the equation,
Figure FDA00027429782700000111
to initialize the covariance matrix.
3. The method of claim 1, wherein the method for predicting the lifetime of a lithium ion battery based on DAUPF comprises: the Step3 is specifically as follows: calculating a Sigma Point set of 2n +1 sample points
Figure FDA00027429782700000112
Wherein, the point set
Figure FDA00027429782700000113
From point Xk-1And
Figure FDA0002742978270000021
the composition of the components, wherein,
Figure FDA0002742978270000022
as a scaling function.
4. The method of claim 1, wherein the method for predicting the lifetime of a lithium ion battery based on DAUPF comprises: the Step4 is specifically as follows:
Figure FDA0002742978270000023
adding two adaptive factors, wherein the initial values of the two adaptive factors are 1; the first adaptive factor is
Figure FDA0002742978270000024
Wherein Z isk-1Is the observed value of the previous sampling point,
Figure FDA0002742978270000025
the mean value of the observed prediction obtained at Step7 in the previous cycle; the second adaptive factor is
Figure FDA0002742978270000026
Wherein Z isk-1For true value of the system, Zupfk-1And (4) obtaining a predicted value after the Step13 DAUPF algorithm finishes the previous cycle.
5. The method of claim 1, wherein the method for predicting the lifetime of a lithium ion battery based on DAUPF comprises: the mean value of Step5 is
Figure FDA0002742978270000027
By
Figure FDA0002742978270000028
And calculating to obtain the result, wherein,
Figure FDA0002742978270000029
substituting the Sigma point set obtained in Step3 into a nonlinear transformation function to obtain a product; covariance
Figure FDA00027429782700000210
Wherein
Figure FDA00027429782700000211
Wherein λ ═ α2(n+κ)-n；α＝1；ρ＝0；κ＝2。
6. The method of claim 1, wherein the method for predicting the lifetime of a lithium ion battery based on DAUPF comprises: step6 is the second UT transform, generating a new Sigma point set,
Figure FDA00027429782700000212
wherein, the point set
Figure FDA00027429782700000213
By point
Figure FDA00027429782700000214
And
Figure FDA00027429782700000215
composition of, wherein
Figure FDA00027429782700000216
The mean value obtained in Step5.
7. The method of claim 1, wherein the method for predicting the lifetime of a lithium ion battery based on DAUPF comprises: the Step7 observed prediction value
Figure FDA00027429782700000217
New Sigma Point set from Step6
Figure FDA00027429782700000218
Substituting the state equation function to obtain; new observed predicted mean
Figure FDA00027429782700000219
Prediction of values from observations
Figure FDA00027429782700000220
Obtaining the weight; new mean value
Figure FDA00027429782700000221
Weighting the new observation prediction mean value and the observation prediction value to obtain; new covariance
Figure FDA00027429782700000222
New Sigma Point set from Step6
Figure FDA00027429782700000223
Mean values obtained in Stap5
Figure FDA00027429782700000224
And weighting the new observation prediction mean value and the observation prediction value.
8. The method of claim 1, wherein the method for predicting the lifetime of a lithium ion battery based on DAUPF comprises: the Step8 calculates Kalman gain KkNew mean value in Step7
Figure FDA00027429782700000225
With new covariance
Figure FDA00027429782700000226
The product of the inverse matrices; updated system covariance
Figure FDA00027429782700000227
By KaerMangan gain KkWith the new mean value
Figure FDA00027429782700000228
Calculating to obtain; updated state
Figure FDA00027429782700000229
From the new Sigma point set in step6
Figure FDA00027429782700000230
Kalman gain KkObserved value Z of current sampling pointkWith new observed predicted mean
Figure FDA00027429782700000231
The difference of (a) is calculated.
9. The method of claim 1, wherein the method for predicting the lifetime of a lithium ion battery based on DAUPF comprises: the Step13 obtains a predicted value, namely ZupfkAnd substituting the normalized weight value into the state equation function to obtain the normalized weight value, and updating a second self-adaptive factor.
CN201910241178.8A 2019-03-28 2019-03-28 Lithium ion battery life prediction method based on DAUPF Expired - Fee Related CN109917292B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910241178.8A CN109917292B (en) 2019-03-28 2019-03-28 Lithium ion battery life prediction method based on DAUPF

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910241178.8A CN109917292B (en) 2019-03-28 2019-03-28 Lithium ion battery life prediction method based on DAUPF

Publications (2)

Publication Number Publication Date
CN109917292A CN109917292A (en) 2019-06-21
CN109917292B true CN109917292B (en) 2020-12-15

Family

ID=66967283

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910241178.8A Expired - Fee Related CN109917292B (en) 2019-03-28 2019-03-28 Lithium ion battery life prediction method based on DAUPF

Country Status (1)

Country Link
CN (1) CN109917292B (en)

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110492866B (en) * 2019-07-22 2022-12-09 航天东方红卫星有限公司 Kalman filtering method for moving target
CN110442941B (en) * 2019-07-25 2022-04-29 桂林电子科技大学 Battery state and RUL prediction method based on particle filtering and process noise fusion
CN111680848A (en) * 2020-07-27 2020-09-18 中南大学 Battery life prediction method based on prediction model fusion and storage medium
CN112285568B (en) * 2020-10-21 2023-11-14 合肥工业大学 Method for estimating residual discharge time based on energy state of power lithium battery
CN112560916B (en) * 2020-12-09 2022-11-01 甘肃靖远航天风力发电有限公司 Wind power tower barrel overturning intelligent diagnosis method based on tilt angle sensor information
CN112763929B (en) * 2020-12-31 2024-03-08 华东理工大学 Method and device for predicting health of battery monomer of energy storage power station system
CN114791993B (en) * 2022-05-16 2022-11-11 江苏大学 Power battery pack SOH prediction method and system

Family Cites Families (22)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6064857A (en) * 1997-04-15 2000-05-16 Globalstar L.P. Dual mode satellite telephone with hybrid battery/capacitor power supply
US6310789B1 (en) * 1999-06-25 2001-10-30 The Procter & Gamble Company Dynamically-controlled, intrinsically regulated charge pump power converter
US7612532B2 (en) * 2005-06-21 2009-11-03 Gm Global Technology Operations, Inc. Method for controlling and monitoring using a state estimator having variable forgetting factors
JP2009207332A (en) * 2008-02-29 2009-09-10 Techno Core International Kk Charger apparatus for pack battery, and quality decision apparatus for pack battery
US8190384B2 (en) * 2011-10-27 2012-05-29 Sakti3, Inc. Method and system for operating a battery in a selected application
KR20150047873A (en) * 2013-10-25 2015-05-06 주식회사 엘지화학 Non-aqueous electrolyte solution for lithium secondary battery and lithium secondary battery comprising the same
CN103675706B (en) * 2013-12-13 2016-04-13 桂林电子科技大学 A kind of power battery electric charge quantity estimation method
CN104267261B (en) * 2014-10-29 2017-02-15 哈尔滨工业大学 On-line secondary battery simplified impedance spectroscopy model parameter estimating method based on fractional order united Kalman filtering
CN104502851A (en) * 2014-12-12 2015-04-08 广西科技大学 SOC (Stress Optical Coefficient) estimation method based on AUKF (Adaptive Unscented Kalman Filter) algorithm
CN105277896B (en) * 2015-10-26 2018-01-26 安徽理工大学 Lithium battery method for predicting residual useful life based on ELM MUKF
CN105629175A (en) * 2015-12-29 2016-06-01 北京航天测控技术有限公司 Lithium ion battery life prediction method based on unscented Kalman filtering (UKF)
US10686321B2 (en) * 2016-01-29 2020-06-16 Robert Bosch Gmbh Secondary battery management
CN105974329A (en) * 2016-07-22 2016-09-28 深圳市沃特玛电池有限公司 Method for estimating SOH of battery pack
CN107664751A (en) * 2016-07-28 2018-02-06 中兴通讯股份有限公司 The measuring method and measuring and calculating device of a kind of real-time state-of-charge of battery
US10099679B2 (en) * 2016-10-20 2018-10-16 Fca Us Llc Battery state and parameter estimation using a mixed sigma-point kalman filtering and recursive least squares technique
CN108693472B (en) * 2017-04-12 2020-11-10 上海蓝诺新能源技术有限公司 Battery equivalent model parameter online identification method
CN107153163A (en) * 2017-06-20 2017-09-12 福建工程学院 A kind of lithium battery SOC estimation method based on adaptive UKF
CN107387064A (en) * 2017-07-27 2017-11-24 河南科技学院 A kind of new explosive-removal robot tunnel enters localization method
CN108875126A (en) * 2018-04-27 2018-11-23 中国航空无线电电子研究所 Electrolytic capacitor method for predicting residual useful life
CN108872870A (en) * 2018-06-21 2018-11-23 浙江工业大学 A kind of lithium battery SOC estimation method based on particle group optimizing expanded Kalman filtration algorithm
CN108594135A (en) * 2018-06-28 2018-09-28 南京理工大学 A kind of SOC estimation method for the control of lithium battery balance charge/discharge
CN109444757A (en) * 2018-10-09 2019-03-08 杭州中恒云能源互联网技术有限公司 A kind of residual capacity of power battery of electric automobile evaluation method

Also Published As

Publication number Publication date
CN109917292A (en) 2019-06-21

Similar Documents

Publication Publication Date Title
CN109917292B (en) Lithium ion battery life prediction method based on DAUPF
Kim et al. Data-efficient parameter identification of electrochemical lithium-ion battery model using deep Bayesian harmony search
Li et al. A new prognostics method for state of health estimation of lithium-ion batteries based on a mixture of Gaussian process models and particle filter
Chen et al. State-of-charge estimation of lithium-ion battery using an improved neural network model and extended Kalman filter
Li et al. A multi-model probability SOC fusion estimation approach using an improved adaptive unscented Kalman filter technique
Chen et al. Battery state-of-health estimation based on a metabolic extreme learning machine combining degradation state model and error compensation
Chen et al. An empirical-data hybrid driven approach for remaining useful life prediction of lithium-ion batteries considering capacity diving
Walker et al. Comparison of a particle filter and other state estimation methods for prognostics of lithium-ion batteries
Shen et al. Accurate state of charge estimation with model mismatch for li-ion batteries: A joint moving horizon estimation approach
Zheng et al. Novel methods for estimating lithium-ion battery state of energy and maximum available energy
Liu et al. Prognostics for state of health estimation of lithium-ion batteries based on combination Gaussian process functional regression
Lai et al. A novel method for state of energy estimation of lithium-ion batteries using particle filter and extended Kalman filter
Cheng et al. Residual lifetime prediction for lithium-ion battery based on functional principal component analysis and Bayesian approach
Ren et al. A comparative study of the influence of different open circuit voltage tests on model‐based state of charge estimation for lithium‐ion batteries
Ma et al. Robust state of charge estimation for Li-ion batteries based on cubature kalman filter with generalized maximum correntropy criterion
Lai et al. A joint state-of-health and state-of-energy estimation method for lithium-ion batteries through combining the forgetting factor recursive least squares and unscented Kalman filter
Mussi et al. A voltage dynamic-based state of charge estimation method for batteries storage systems
Venegas et al. Kalman filter and classical Preisach hysteresis model applied to the state of charge battery estimation
CN109633470B (en) Estimation method for battery real-time full charge time based on EKF-GPR and daily segment data
Vedhanayaki et al. Certain investigation and implementation of Coulomb counting based unscented Kalman filter for state of charge estimation of lithium-ion batteries used in electric vehicle application
Dang et al. Multi-kernel correntropy based extended Kalman filtering for state-of-charge estimation
Takyi-Aninakwa et al. An ASTSEKF optimizer with nonlinear condition adaptability for accurate SOC estimation of lithium-ion batteries
Liu et al. Online state of charge estimation for lithium-ion battery by combining incremental autoregressive and moving average modeling with adaptive H-infinity filter
Alhalali et al. State estimator for electrical distribution systems based on a particle filter
Lashgari et al. State of Health analysis for Lithium-ion Batteries considering temperature effect

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
GR01 Patent grant
GR01 Patent grant
TR01 Transfer of patent right

Effective date of registration: 20211210

Address after: 532605 Building D, yizilian mahogany processing zone, No. 8, Youyiguan Industrial Park, Xiashi Town, Pingxiang City, Chongzuo City, Guangxi Zhuang Autonomous Region

Patentee after: Guangxi Jubang Energy Co.,Ltd.

Address before: 100089 Beijing Haidian District 2 Huayuan Road peony Venture Building 4 floor 1424 room

Patentee before: Beijing Zhonglian Technology Service Co.,Ltd.

Effective date of registration: 20211210

Address after: 100089 Beijing Haidian District 2 Huayuan Road peony Venture Building 4 floor 1424 room

Patentee after: Beijing Zhonglian Technology Service Co.,Ltd.

Address before: No.56, Xisanhuan North Road, Haidian District, Beijing 100048

Patentee before: Capital Normal University

TR01 Transfer of patent right
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20201215

CF01 Termination of patent right due to non-payment of annual fee