CN111650517A - Battery state of charge estimation method - Google Patents

Battery state of charge estimation method Download PDF

Info

Publication number
CN111650517A
CN111650517A CN202010309691.9A CN202010309691A CN111650517A CN 111650517 A CN111650517 A CN 111650517A CN 202010309691 A CN202010309691 A CN 202010309691A CN 111650517 A CN111650517 A CN 111650517A
Authority
CN
China
Prior art keywords
equations
state
value
battery
algorithm
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.)
Granted
Application number
CN202010309691.9A
Other languages
Chinese (zh)
Other versions
CN111650517B (en
Inventor
王伟达
杨超
项昌乐
刘金刚
韩立金
倪俊
张中国
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 Institute of Technology BIT
Original Assignee
Beijing Institute of Technology BIT
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 Beijing Institute of Technology BIT filed Critical Beijing Institute of Technology BIT
Priority to CN202010309691.9A priority Critical patent/CN111650517B/en
Publication of CN111650517A publication Critical patent/CN111650517A/en
Application granted granted Critical
Publication of CN111650517B publication Critical patent/CN111650517B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Tests Of Electric Status Of Batteries (AREA)

Abstract

The invention discloses a battery state of charge estimation method, which aims at providing a nonlinear state relation existing in a battery model, provides a state of charge SOC estimation algorithm based on STF-SRUKF, combines a strong tracking filter theory STF algorithm with a square root unscented Kalman filter algorithm SR-UKF algorithm, on one hand, uses a covariance square root to replace a covariance matrix for recursive calculation, and strengthens the stability of iteration of the square root unscented Kalman filter algorithm SR-UKF algorithm, on the other hand, introduces a time-varying fading factor based on the idea of strong tracking filtering, and improves the robustness of the square root unscented Kalman filter algorithm SR-UKF algorithm on parameter perturbation, thereby improving the estimation precision of the battery state of charge SOC under the influence of complex working conditions and strong noise.

Description

Battery state of charge estimation method
Technical Field
The invention relates to the technical field of battery state of charge estimation, in particular to a battery state of charge estimation method.
Background
Based on lithium iron phosphate batteries which are widely used in electric vehicles nowadays, as shown in fig. 1. The lithium iron phosphate battery has high safety performance, long service life, good high-temperature performance and no memory effect, and meets the requirement of environmental protection. For a four-wheel hub motor-driven electric vehicle, a power battery pack is a core component of the electric vehicle, and directly determines the driving range and the control mode of the electric vehicle. In order to prevent overcharge and overdischarge of the battery during use, improve the service life of the battery and the drivability of the electric vehicle, it is necessary to estimate the state of charge SOC of the battery.
Disclosure of Invention
The invention aims to provide a SOC estimation algorithm based on STF-SRUKF aiming at a nonlinear state relation existing in a battery model, combine a strong tracking filter theory STF algorithm with a square root unscented Kalman filtering algorithm SR-UKF algorithm, on one hand, use a covariance square root to replace a covariance matrix for recursive calculation, and strengthen the iteration stability of the square root unscented Kalman filtering algorithm SR-UKF algorithm, on the other hand, introduce a time-varying fading factor based on the idea of strong tracking filtering, and improve the robustness of the square root unscented Kalman filtering algorithm SR-UKF algorithm on parameter perturbation, thereby improving the estimation precision of the battery SOC under the influence of complex working conditions and strong noise.
In order to solve the technical problems, the invention provides the following technical scheme:
a battery state of charge estimation method, comprising the steps of:
step 1, establishing a first-order RC model of the battery:
according to the circuit principle, a characteristic equation of a first-order RC model is deduced as shown in a formula 3-1:
Figure RE-GDA0002595659000000011
the current is used as an input value of a system, the end circuit voltage is used as a measured value of the system, the battery SOC and the polarization voltage are used as state quantities of the system, and a state equation of the power battery system can be established as follows, wherein the equation is shown in a formula 3-2:
Figure RE-GDA0002595659000000012
wherein, the polarization capacitance C of the battery at the k momentpPolarization resistance RpAnd open circuit voltage UOCAll are SOC correlation functions, and an experimental method is needed for identification and fitting; open circuit voltage adopts OCV test to carry out point taking experiment; the remaining parameters of the cell model, e.g. polarization capacitance CpPolarization resistance RpThe time constant tau needs HPPC experiment to identify;
step 2, calculating the state estimation value of the system at the k moment according to the formula 3-2
Figure RE-GDA0002595659000000021
And PkEstimating the state of the system at the moment k +1 according to the data;
for the Kalman filtering recursion system, the state estimation steps are as follows, and the formula is 3-3:
equation 3-3:
Figure RE-GDA0002595659000000022
wherein
Figure RE-GDA0002595659000000023
Is a residual sequence obtained by a state estimation filtering equation; under the condition that the Kalman filtering theory meets the formula 3-4, the strong tracking filter is added with the formula 3-5, so that residual sequences at different moments are orthogonal everywhere;
equations 3-4:
Figure RE-GDA0002595659000000024
equations 3-5:
Figure RE-GDA0002595659000000025
in order to establish the formulas 3-5, the STF algorithm introduces a time-varying fading factor lambda, and adjusts the prediction error covariance matrix in real time, so as to further update the Kalman gain; the calculation method of the fading factor λ is as follows:
the formula: 3-6:
Figure RE-GDA0002595659000000026
equations 3-7:
Figure RE-GDA0002595659000000027
equations 3-8:
Figure RE-GDA0002595659000000028
equations 3-9:
Figure RE-GDA0002595659000000029
wherein, VkFor the residual covariance matrix, the following is defined, equations 3-10:
formula 3-10:
Figure RE-GDA00025956590000000210
in the formula, rho is more than 0 and less than or equal to 1 and is a forgetting factor, beta is more than or equal to 1 and is a weakening factor, and the estimation result can be smoother by increasing the value of beta; f and H are Jacobian matrixes of a system state equation and an observation equation respectively;
constructing sigma points according to the mean of the state vector x
Figure RE-GDA00025956590000000212
With covariance P, the sigma point set { ξ is chosen as followsi1,2, …, and a weight W corresponding to the point seti (m)And Wi (c)Equation 3-11:
equations 3-11:
Figure RE-GDA00025956590000000211
where κ is a scale condition coefficient, determining the samplingPrecision of the value, sigma point set ξ at this timei-approximation of the gaussian distribution of x;
the sigma point set is propagated through a nonlinear function, the selected sigma point set is calculated through a nonlinear function f (x), and the nonlinear propagated sigma point set { Y) is obtainediWhen the probability distribution of Y is close to the probability distribution of the nonlinear function calculation value Y ═ f (x), equations 3-12:
formulas 3 to 12: y isi＝f(χi),i＝0,1,...,2n
Calculating the mean value and covariance value of the transformed point set, and performing sigma point set { Y after transformationiWeighting to obtain the mean value and covariance of the output variables by calculation, wherein the weight value is equal to that of the input variables:
equations 3-13:
Figure RE-GDA0002595659000000031
equations 3-14:
Figure RE-GDA0002595659000000032
wherein, Wi mAnd Wi cThe weighting coefficients used for calculating the mean and covariance of the y-values are:
Figure RE-GDA0002595659000000033
equations 3-15:
Figure RE-GDA0002595659000000034
wherein κ is α2The value of (n + lambda) -n, α is positive and is usually slightly less than 1, β is a state distribution parameter, the parameter is generally 2 for a Gaussian distribution system, lambda is usually set to be 0 or 3-n, the precision of variance transformation can be improved by adjusting the value of β, and the precision of the estimated mean value can be improved by adjusting the values of α and lambda;
for nonlinear systems, equations 3-16:
formulas 3 to 16:
Figure RE-GDA0002595659000000035
the iteration steps of the square root unscented kalman filter are as follows:
initialization:
equations 3-17:
Figure RE-GDA0002595659000000036
calculate Sigma point:
equations 3-18:
Figure RE-GDA0002595659000000037
and step 3, updating the time as follows:
equations 3-19:
Figure RE-GDA0002595659000000038
formulas 3 to 20:
Figure RE-GDA0002595659000000039
equations 3-21:
Figure RE-GDA00025956590000000310
formulas 3 to 22:
Figure RE-GDA00025956590000000311
equations 3-23: y isk/k-1＝H[χk/k-1]
Formulas 3 to 24:
Figure RE-GDA0002595659000000041
wherein QR represents QR decomposition, and chloripdate represents Cholesky first-order update, which are matlab executable function statements;
the measurement is updated as follows:
formulas 3 to 25:
Figure RE-GDA0002595659000000042
equations 3-26:
Figure RE-GDA0002595659000000043
step 4, calculating an fading factor:
equations 3-27:
Figure RE-GDA0002595659000000044
equations 3-28:
Figure RE-GDA0002595659000000045
formulas 3 to 29:
Figure RE-GDA0002595659000000046
equations 3-30:
Figure RE-GDA0002595659000000047
equations 3-31: sk＝cholupdate(Sk/k-1,U,-1)
Because SR-UKF also needs choleupdate in the iterative process, and the product of the column vector of U and its transposition in the formula 3-31 may have non-positive character, the square root S of covariance P after choleupdate also loses positive character; to avoid this, the following modifications are made to equations 3-31:
let R be U (: i)TIf R is a positive definite matrix, then equations 3-32:
Sk＝cholupdate(Sk/k-1,U(:,i),-1)
otherwise, equations 3-33:
Figure RE-GDA0002595659000000048
similarly, equations 3-22 and equations 3-26 are modified as follows:
if sign (W)1 (c)) -1, then:
equations 3-34:
Figure RE-GDA0002595659000000049
equations 3-35:
Figure RE-GDA00025956590000000410
if not, then,
formulas 3 to 36:
Figure RE-GDA00025956590000000411
equations 3-37:
Figure RE-GDA00025956590000000412
step 5, calculating the square root of the covariance matrix of the state prediction error:
equations 3-38:
Figure RE-GDA00025956590000000413
equations 3-39:
Figure RE-GDA0002595659000000051
step 6, updating the system state and the square root of the covariance according to the formulas 3-27 to 3-31, namely the estimated value at the k +1 moment
Figure RE-GDA0002595659000000052
Including the estimated value of the battery SOC at the current moment and the polarization voltage UpAn estimate of the current time.
Compared with the prior art, the invention has the following beneficial effects: the method is characterized in that a first-order RC model is used for modeling the power battery of the electric automobile, the parameters of the model are identified, and then a strong tracking filter theory (STF) and a square root unscented Kalman filter algorithm (SR-UKF) are combined, so that the stability of the filter algorithm is ensured, the tracking capability of the algorithm on emergency is improved, and the SOC of the battery is estimated based on the algorithm. And performing off-line simulation verification by using the acquired battery voltage and current data, and substituting three different estimation algorithms for result comparison. Simulation results show that stable, effective and accurate estimation can be performed on the SOC of the battery under the condition of large noise interference based on the STF-SRUKF algorithm.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention. In the drawings:
fig. 1 is a schematic diagram of a lithium iron phosphate battery of an electric vehicle in the prior art.
Fig. 2 is a schematic diagram of an RC battery model in an embodiment of the invention.
Fig. 3 is a graph illustrating the relationship between the open-circuit voltage and the state of charge SOC of a battery cell according to an embodiment of the present invention.
FIG. 4 is a diagram comparing a battery simulation model with experimental data in an embodiment of the present invention.
FIG. 5 is a diagram illustrating a comparison between EKF algorithm and UT transformation error in an embodiment of the present invention.
Fig. 6 is a schematic diagram of a voltage signal with noise added as an input of a battery state of charge SOC estimation algorithm according to an embodiment of the present invention.
Fig. 7 is a schematic diagram of a current signal with noise added as an input of a battery state of charge SOC estimation algorithm according to an embodiment of the present invention.
Fig. 8 is a schematic diagram of a battery SOC estimation result according to an embodiment of the present invention.
FIG. 9 is a flow chart of a method in an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in FIG. 1, the allowable working temperature range of the iron phosphate lithium battery is-30 to 55 degrees. Because the electrochemical reaction in the battery is very complicated, the parameters of the battery can be changed under the influence of various external factors such as temperature, pressure and the like. Therefore, the robustness requirement for the battery state of charge SOC estimation algorithm is high. The following table shows the performance parameters of the lithium iron phosphate battery of fig. 1.
Figure RE-GDA0002595659000000061
The invention provides a method for estimating the state of charge of a battery, which adopts a first-order RC battery model, also called Thevenin model, as shown in figure 2, wherein the model simulates the polarization phenomenon of the battery by using a resistor/capacitor parallel circuit and describes the internal resistance of the battery by using a resistor. Both of these parameters change as the state of charge, SOC, of the battery changes.
According to the circuit principle, the characteristic equation of a first-order RC model can be deduced to be formula 3-1:
Figure RE-GDA0002595659000000062
the current is used as an input value of a system, the end circuit voltage is used as a measured value of the system, the battery SOC and the polarization voltage are used as state quantities of the system, and a state equation of the power battery system can be established as follows, wherein the equation is shown in a formula 3-2:
Figure RE-GDA0002595659000000063
wherein, the polarization capacitance C of the battery at the k momentpPolarization resistance RpAnd open circuit voltage UOCAre all SOC related functions, and need to be identified and fitted by an experimental method. The open circuit voltage is subjected to a point-taking experiment by using an OCV test, and experimental data and a fitting curve are shown in FIG. 3.
Battery modelOther parameters of e.g. polarization capacitance CpPolarization resistance RpThe time constant τ is identified by HPPC experiments. Because the conventional HPPC experiment only performs 0.75C charging and 1C discharging, and for the vehicle power battery, under the working conditions of vehicle braking energy recovery and rapid acceleration, the short-time charging and discharging of the battery can reach 5C, and under the working conditions, if the conventional HPPC experiment mode is adopted, the charging and discharging behaviors of the battery cannot be accurately modeled. Therefore, the charging current and the discharging current of the HPPC experiment are respectively 3.75C and 5C, 10 points are sequentially taken from 100% to 10% according to the SOC, and the interval between each point is 10%.
According to current and voltage data acquired by experiments, combining a battery model state equation and performing off-line identification based on a least square method, thereby obtaining a polarization capacitor CpPolarization resistance RpWith respect to the time constant τ and SOC. A first-order RC model of the battery is established in Simulink based on the identification result, the simulation result is compared with the experimental data, the result is shown in figure 4, and it can be seen that in figure 4, the error between the simulation model result and the experimental result is small, so that the model precision is available.
First, strong tracking filter theory (STF):
the strong tracking filtering theory is firstly proposed by professor Zhoutowa university of Qinghua in 1990, and the core idea is to introduce a time-varying fading factor, correct a state prediction error covariance matrix and a corresponding Kalman gain matrix in a Kalman filtering recursion process, and force a residual sequence to be orthogonal or approximately orthogonal. When the model or the measured value has uncertainty or mutation, the STF algorithm calculates an fading factor for ensuring the irrelevance of the innovation sequence, thereby weakening the influence of historical data on the current filtering calculation value and enabling the algorithm to have the capability of tracking the mutation state.
For the Kalman filtering recursion system, the state estimation steps are as follows, and the formula is 3-3:
equation 3-3:
Figure RE-GDA0002595659000000071
wherein
Figure RE-GDA0002595659000000072
Is the residual sequence found by the state estimation filter equation. Under the condition that the Kalman filtering theory meets the formula 3-4, the strong tracking filter is added with the formula 3-5, so that residual sequences at different moments are orthogonal everywhere.
Equations 3-4:
Figure RE-GDA0002595659000000073
equations 3-5:
Figure RE-GDA0002595659000000074
in order to make equations 3-5 hold, the STF algorithm introduces a time-varying fading factor λ, adjusting the prediction error covariance matrix in real time, and further updating the kalman gain. The calculation method of the fading factor λ is as follows:
the formula: 3-6:
Figure RE-GDA0002595659000000075
equations 3-7:
Figure RE-GDA0002595659000000076
equations 3-8:
Figure RE-GDA0002595659000000077
equations 3-9:
Figure RE-GDA0002595659000000078
wherein, VkFor the residual covariance matrix, the following is defined, equations 3-10:
formula 3-10:
Figure RE-GDA0002595659000000079
in the formula, rho is more than 0 and less than or equal to 1 and is a forgetting factor, beta is more than or equal to 1 and is a weakening factor, and the estimation result can be smoother by increasing the value of beta. F and H are Jacobian matrixes of a system state equation and an observation equation respectively.
II, square root unscented Kalman Filter (SR-UKF):
the extended Kalman filtering adopts Taylor expansion to carry out linearization, and high-order nonlinear terms above the second order are ignored. Therefore, when the degree of non-linearity of the system is high, the accuracy of the EKF algorithm is poor, and the calculation is complex and easy to make errors. In order to solve the problem of loss of precision of linearization of the EKF algorithm, Julian professor of the Oxford university proposes a non-linear Kalman filtering algorithm based on UT transformation in 1995, omits the process of solving the Jacobian matrix, and approximates the statistical characteristics of the state quantities by using a series of sigma sampling points, namely the mean and covariance distribution of the sampling points is equal to the mean and covariance distribution of the state quantities. And then carrying out nonlinear transformation on the group of sampling points according to a state space equation of the system, and taking weighted average on the transformed sampling points to obtain the statistical characteristic of the approximate variable after nonlinear transformation, wherein the statistical characteristic is also equal to the mean value and the covariance of the state quantity.
The core method of Unscented Kalman Filtering (UKF) is UT transform. Intuitively, it is easier to approximate a probability distribution than to approximate an arbitrary non-linear function. Assume the mean of the state vector x
Figure RE-GDA0002595659000000081
And covariance P is known, a set of determined vectors is found, called sigma points, with mean and covariance of the set of sigma points as
Figure RE-GDA0002595659000000082
And P, then each determined vector is subjected to a known nonlinear function f (x) and h (x) to obtain a transformed vector, namely the sigma point is substituted into the nonlinear function for calculation. The mean value and covariance of the transformed vector can accurately approximate the real mean value and covariance of the original state vector after nonlinear function transformation. The UT transform can achieve the fourth order accuracy of a taylor series, as shown in fig. 5.
The specific steps of the UT conversion algorithm are as follows:
(1) constructing sigma points
According to the mean value of the state vector x
Figure RE-GDA0002595659000000087
With covariance P, the sigma point set { ξ is chosen as followsi1,2, …, and a weight W corresponding to the point seti (m)And Wi (c)Equation 3-11:
equations 3-11:
Figure RE-GDA0002595659000000083
wherein, kappa is a scale condition coefficient and determines the precision of the sampling value, and the sigma point set { ξ in the momentiThe gaussian distribution of x is approximated.
(2) The sigma point set is propagated by a non-linear function
Calculating the selected sigma point set through a nonlinear function f (x) to obtain a nonlinear propagated sigma point set { Y }iWhen the probability distribution of Y is close to the probability distribution of the nonlinear function calculation value Y ═ f (x), equations 3-12:
formulas 3 to 12: y isi＝f(χi),i＝0,1,...,2n
(3) Mean and covariance value calculation of transformed point sets
For the transformed sigma point set { YiWeighting to obtain the mean value and covariance of the output variables by calculation, wherein the weight value is equal to that of the input variables:
equations 3-13:
Figure RE-GDA0002595659000000084
equations 3-14:
Figure RE-GDA0002595659000000085
wherein, Wi mAnd Wi cThe weighting coefficients used for calculating the mean and covariance of the y-values are:
Figure RE-GDA0002595659000000091
equations 3-15:
Figure RE-GDA0002595659000000092
wherein κ is α2The (n + λ) -n, α values are positive, typically slightly less than 1.β is the state distribution parameter, for gaussian systems this parameter is typically 2. λ is typically set to 0 or 3-n. adjusting the β value improves the variance transformation accuracy, and adjusting the α and λ values improves the mean estimate accuracy.
When calculating the Sigma point, the UKF algorithm needs Cholesky decomposition on the covariance matrix, that is, positive determination of the covariance matrix is required. In the computer calculation process, truncation errors can be generated due to the limitation of calculation accuracy. With the accumulation of truncation errors or stepwise changes of parameters, the covariance matrix will likely lose the positive nature, thereby making the filtering process impossible. The square root UKF algorithm iteratively propagates the square root matrix of the covariance matrix, thereby avoiding factorization in the iterative process.
For nonlinear systems, equations 3-16:
formulas 3 to 16:
Figure RE-GDA0002595659000000093
the iteration steps of the square root unscented kalman filter are as follows:
initialization:
equations 3-17:
Figure RE-GDA0002595659000000094
calculate Sigma point:
equations 3-18:
Figure RE-GDA0002595659000000095
the time is updated as:
equations 3-19:
Figure RE-GDA0002595659000000096
formulas 3 to 20:
Figure RE-GDA0002595659000000097
equations 3-21:
Figure RE-GDA0002595659000000098
formulas 3 to 22:
Figure RE-GDA0002595659000000099
equations 3-23:
Figure RE-GDA00025956590000000910
formulas 3 to 24:
Figure RE-GDA00025956590000000911
wherein QR represents QR decomposition, and chloripdate represents Cholesky first-order update, which are matlab executable function statements.
The measurement is updated as follows:
formulas 3 to 25:
Figure RE-GDA00025956590000000912
equations 3-26:
Figure RE-GDA0002595659000000101
equations 3-27:
Figure RE-GDA0002595659000000102
equations 3-28:
Figure RE-GDA0002595659000000103
formulas 3 to 29:
Figure RE-GDA0002595659000000104
equations 3-30:
Figure RE-GDA0002595659000000105
equations 3-31: sk＝cholupdate(Sk/k-1,U,-1)
Since SR-UKF also needs to carry out choleupdate decomposition in the iterative process, the product of the column vector of U and its transpose in the formulas 3-31 may have non-positive character, and the square root S of covariance P after choleupdate decomposition also loses positive character. To avoid this, the following modifications are made to equations 3-31:
let R be U (: i)TIf R is a positive definite matrix, then equations 3-32:
Sk＝cholupdate(Sk/k-1,U(:,i),-1)
otherwise, equations 3-33:
Figure RE-GDA0002595659000000106
similarly, equations 3-22 and equations 3-26 are modified as follows:
if sign (W)1 (c)) -1, then equation 3-34:
equations 3-34:
Figure RE-GDA0002595659000000107
equations 3-35:
Figure RE-GDA0002595659000000108
otherwise, equations 3-36:
formulas 3 to 36:
Figure RE-GDA0002595659000000109
equations 3-37:
Figure RE-GDA00025956590000001010
thirdly, STF-SRUKF algorithm:
in summary, as shown in fig. 9, the present invention includes the following steps:
step 1, establishing a first-order RC model of the battery: the state space equation of the SOC estimation system of the battery can be known by the formula 3-2 and is substituted into the STF-SRUKF algorithm;
step 2, calculating a state estimation value of the system at the k moment according to a system state equation and an initial value
Figure RE-GDA00025956590000001012
And PkEstimating the state of the system at the moment k +1 according to the data;
and step 3, time updating: the state estimation and innovation covariance matrices for the system are calculated from equations 3-19 through equations 3-26. At the moment, a time-varying fading factor is not introduced into the prediction covariance matrix;
and 4, step 4: calculating an fading factor: obtaining a lambda value at the k +1 moment according to the formulas 3-6 to 3-10;
step 5, calculating the square root of the covariance matrix of the state prediction error: equations 3-25 and equations 3-26 are modified to:
equations 3-38:
Figure RE-GDA00025956590000001013
equations 3-39:
Figure RE-GDA00025956590000001011
step 6: the system state and the square root of the covariance are updated according to equations 3-27 through 3-31, i.e., the estimated value at time k +1
Figure RE-GDA0002595659000000111
Including the estimated value of the battery SOC at the current moment and the polarization voltage UpAn estimate of the current time.
In order to verify the effect and the anti-interference capability of the battery SOC estimation method, Gaussian white noise interference with a mean value of zero and a variance of 1A is added to current signals collected by experimental equipment, Gaussian white noise interference with a mean value of zero and a variance of 50mV is added to collected voltage signals, and the voltage and current signals added with noise are used as the input of a battery SOC estimation algorithm, as shown in FIG. 6 and FIG. 7.
The traditional EKF algorithm, the original SR-UKF algorithm and the STF-SRUKF algorithm proposed in the chapter are respectively adopted to estimate the SOC of the battery, the results of FIG. 8 show that the STF-SRUKF algorithm proposed in the chapter can remarkably reduce the influence of measurement noise on the estimation of the SOC of the battery, the precision is greatly improved compared with the EKF algorithm and the SR-UKF algorithm, the estimation precision of the STF-SRUKF algorithm is the highest under the larger noise interference, and the error is within +/-0.07.
The error analysis is shown in the following table 2, and it can be seen that, no matter the Root Mean Square Error (RMSE) or the Maximum Error (ME), the estimation result of the STF-SRUKF algorithm provided by the invention on the battery SOC is greatly superior to the traditional EKF algorithm and the original SR-UKF algorithm.
TABLE 2 Battery SOC estimation result error
Figure RE-GDA0002595659000000112
The method uses the first-order RC model to model the power battery of the electric automobile, identifies the parameters of the model, and then combines the strong tracking filter theory (STF) with the square root unscented Kalman filter algorithm (SR-UKF), thereby not only ensuring the stability of the filter algorithm, but also improving the tracking capability of the algorithm to emergencies, and estimating the SOC of the battery based on the algorithm. And performing off-line simulation verification by using the acquired battery voltage and current data, and substituting three different estimation algorithms for result comparison. Simulation results show that stable, effective and accurate estimation can be performed on the SOC of the battery under the condition of large noise interference based on the STF-SRUKF algorithm.
Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that changes may be made in the embodiments and/or equivalents thereof without departing from the spirit and scope of the invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (1)

1. A method of estimating state of charge of a battery, comprising the steps of:
step 1, establishing a first-order RC model of the battery:
according to the circuit principle, a characteristic equation of a first-order RC model is deduced as shown in a formula 3-1:
Figure FDA0002457179290000011
the current is used as an input value of a system, the end circuit voltage is used as a measured value of the system, the battery SOC and the polarization voltage are used as state quantities of the system, and a state equation of the power battery system can be established as follows, wherein the equation is shown in a formula 3-2:
Figure FDA0002457179290000012
wherein, the polarization capacitance C of the battery at the k momentpPolarization resistance RpAnd open circuit voltage UOCAll are SOC correlation functions, and an experimental method is needed for identification and fitting; open circuit voltage adopts OCV test to carry out point taking experiment; the remaining parameters of the cell model, e.g. polarization capacitance CpPolarization resistance RpThe time constant tau needs HPPC experiment to identify;
step 2, calculating the state estimation value of the system at the k moment according to the formula 3-2
Figure FDA0002457179290000013
And PkBased on this estimateCounting the state of the system at the time k + 1;
for the Kalman filtering recursion system, the state estimation steps are as follows, and the formula is 3-3:
equation 3-3:
Figure FDA0002457179290000014
wherein
Figure FDA0002457179290000015
Is a residual sequence obtained by a state estimation filtering equation; under the condition that the Kalman filtering theory meets the formula 3-4, the strong tracking filter is added with the formula 3-5, so that residual sequences at different moments are orthogonal everywhere;
equations 3-4:
Figure FDA0002457179290000016
equations 3-5:
Figure FDA0002457179290000017
in order to establish the formulas 3-5, the STF algorithm introduces a time-varying fading factor lambda, and adjusts the prediction error covariance matrix in real time, so as to further update the Kalman gain; the calculation method of the fading factor λ is as follows:
the formula: 3-6:
Figure FDA0002457179290000018
equations 3-7:
Figure FDA0002457179290000021
equations 3-8:
Figure FDA0002457179290000022
equations 3-9:
Figure FDA0002457179290000023
wherein, VkFor the residual covariance matrix, the following is defined, equations 3-10:
formula 3-10:
Figure FDA0002457179290000024
in the formula, rho is more than 0 and less than or equal to 1 and is a forgetting factor, beta is more than or equal to 1 and is a weakening factor, and the estimation result can be smoother by increasing the value of beta; f and H are Jacobian matrixes of a system state equation and an observation equation respectively;
constructing sigma points according to the mean of the state vector x
Figure FDA0002457179290000029
With covariance P, the sigma point set { ξ is chosen as followsi1,2, …, and a weight W corresponding to the point seti (m)And Wi (c)Equation 3-11:
equations 3-11:
Figure FDA0002457179290000025
wherein, k is a scale condition coefficient, which determines the precision of the sampling value, and the sigma point set { ξ }i-approximation of the gaussian distribution of x;
the sigma point set is propagated through a nonlinear function, the selected sigma point set is calculated through a nonlinear function f (x), and the nonlinear propagated sigma point set { Y) is obtainediWhen the probability distribution of Y is close to the probability distribution of the nonlinear function calculation value Y ═ f (x), equations 3-12:
formulas 3 to 12: y isi＝f(χi),i＝0,1,...,2n
Calculating the mean value and covariance value of the transformed point set, and performing sigma point set { Y after transformationiWeighting to obtain the mean value and covariance of the output variables by calculation, wherein the weight value is equal to that of the input variables:
equations 3-13:
Figure FDA0002457179290000026
equations 3-14:
Figure FDA0002457179290000027
wherein, Wi mAnd Wi cThe weighting coefficients used for calculating the mean and covariance of the y-values are:
equations 3-15:
Figure FDA0002457179290000028
wherein κ is α2The value of (n + lambda) -n, α is positive and is usually slightly less than 1, β is a state distribution parameter, the parameter is generally 2 for a Gaussian distribution system, lambda is usually set to be 0 or 3-n, the precision of variance transformation can be improved by adjusting the value of β, and the precision of the estimated mean value can be improved by adjusting the values of α and lambda;
for nonlinear systems, equations 3-16:
formulas 3 to 16:
Figure FDA0002457179290000031
the iteration steps of the square root unscented kalman filter are as follows:
initialization:
equations 3-17:
Figure FDA0002457179290000032
calculate Sigma point:
equations 3-18:
Figure FDA0002457179290000033
and step 3, updating the time as follows:
equations 3-19: chi shapek/k-1＝F[χk-1]
Formulas 3 to 20:
Figure FDA0002457179290000034
equations 3-21:
Figure FDA0002457179290000035
formulas 3 to 22:
Figure FDA0002457179290000036
equations 3-23: y isk/k-1＝H[χk/k-1]
Formulas 3 to 24:
Figure FDA0002457179290000037
wherein QR represents QR decomposition, and chloripdate represents Cholesky first-order update, which are matlab executable function statements;
the measurement is updated as follows:
formulas 3 to 25:
Figure FDA0002457179290000038
equations 3-26:
Figure FDA0002457179290000039
step 4, calculating an fading factor:
equations 3-27:
Figure FDA00024571792900000310
equations 3-28:
Figure FDA00024571792900000311
formulas 3 to 29:
Figure FDA00024571792900000312
equations 3-30:
Figure FDA00024571792900000313
equations 3-31: sk＝cholupdate(Sk/k-1,U,-1)
Because SR-UKF also needs choleupdate in the iterative process, and the product of the column vector of U and its transposition in the formula 3-31 may have non-positive character, the square root S of covariance P after choleupdate also loses positive character; to avoid this, the following modifications are made to equations 3-31:
let R be U (: i)TIf R is a positive definite matrix, then equations 3-32:
Sk＝cholupdate(Sk/k-1,U(:,i),-1)
otherwise, equations 3-33:
Figure FDA0002457179290000041
similarly, equations 3-22 and equations 3-26 are modified as follows:
if sign (W)1 (c)) -1, then:
equations 3-34:
Figure FDA0002457179290000042
equations 3-35:
Figure FDA0002457179290000043
if not, then,
formulas 3 to 36:
Figure FDA0002457179290000044
equations 3-37:
Figure FDA0002457179290000045
step 5, calculating the square root of the covariance matrix of the state prediction error:
equations 3-38:
Figure FDA0002457179290000046
equations 3-39:
Figure FDA0002457179290000047
step 6, updating the system state and the square root of the covariance according to the formulas 3-27 to 3-31, namely the estimated value at the k +1 moment
Figure FDA0002457179290000048
Including the estimated value of the battery SOC at the current moment and the polarization voltage UpAn estimate of the current time.
CN202010309691.9A 2020-04-20 2020-04-20 Battery state of charge estimation method Active CN111650517B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010309691.9A CN111650517B (en) 2020-04-20 2020-04-20 Battery state of charge estimation method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010309691.9A CN111650517B (en) 2020-04-20 2020-04-20 Battery state of charge estimation method

Publications (2)

Publication Number Publication Date
CN111650517A true CN111650517A (en) 2020-09-11
CN111650517B CN111650517B (en) 2021-08-17

Family

ID=72346472

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010309691.9A Active CN111650517B (en) 2020-04-20 2020-04-20 Battery state of charge estimation method

Country Status (1)

Country Link
CN (1) CN111650517B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112114255A (en) * 2020-09-18 2020-12-22 中国计量大学 Lithium battery state of charge estimation method based on combined ensemble Kalman filtering
CN113075561A (en) * 2021-03-23 2021-07-06 广东技术师范大学 Method, system, equipment and medium for dynamically estimating SOC of lithium ion power battery
CN113805075A (en) * 2021-09-15 2021-12-17 上海电机学院 BCRLS-UKF-based lithium battery state of charge estimation method
CN116203432A (en) * 2023-03-23 2023-06-02 广东工业大学 CSO optimization-based unscented Kalman filtering method for predicting battery state of charge

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102998629A (en) * 2012-12-16 2013-03-27 天津大学 Method for estimating lithium battery charge state
CN105093122A (en) * 2015-07-09 2015-11-25 宁波飞拓电器有限公司 Strong-tracking self-adaptive-SQKF-based SOC estimation method of emergency lamp battery
CN105510829A (en) * 2014-09-29 2016-04-20 山东大学 Novel lithium ion power cell SOC estimation method
CN106324521A (en) * 2016-09-05 2017-01-11 北京理工大学 Power battery system parameter and state of charge joint estimation method
CN107219466A (en) * 2017-06-12 2017-09-29 福建工程学院 A kind of lithium battery SOC estimation method for mixing EKF
JP2018096954A (en) * 2016-12-16 2018-06-21 三菱自動車工業株式会社 Battery state estimation device
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
CN110441694A (en) * 2019-07-23 2019-11-12 哈尔滨工业大学（威海） Based on Multiple fading factor Kalman filtering lithium battery charge state estimation method

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102998629A (en) * 2012-12-16 2013-03-27 天津大学 Method for estimating lithium battery charge state
CN105510829A (en) * 2014-09-29 2016-04-20 山东大学 Novel lithium ion power cell SOC estimation method
CN105093122A (en) * 2015-07-09 2015-11-25 宁波飞拓电器有限公司 Strong-tracking self-adaptive-SQKF-based SOC estimation method of emergency lamp battery
CN106324521A (en) * 2016-09-05 2017-01-11 北京理工大学 Power battery system parameter and state of charge joint estimation method
JP2018096954A (en) * 2016-12-16 2018-06-21 三菱自動車工業株式会社 Battery state estimation device
CN107219466A (en) * 2017-06-12 2017-09-29 福建工程学院 A kind of lithium battery SOC estimation method for mixing EKF
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
CN110441694A (en) * 2019-07-23 2019-11-12 哈尔滨工业大学（威海） Based on Multiple fading factor Kalman filtering lithium battery charge state estimation method

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
BIZHONG XIA等: "A New Method for State of Charge Estimation of Lithium-Ion Battery Based on Strong Tracking Cubature Kalman Filter", 《ENERGIES》 *
DI LI等: "State of charge estimation for LiMn2O4 power battery based on strong tracking sigma point Kalman filter", 《JOURNAL OF POWER SOURCES》 *
XIANGYU CUI等: "A New Method for State of Charge Estimation of Lithium-Ion Batteries Using Square Root Cubature Kalman Filter", 《ENERGIES》 *
张佩等: "基于无迹变换强跟踪滤波的锂离子电池SOC估计", 《数字制造科学》 *
费亚龙等: "基于平方根无迹卡尔曼滤波的锂电池状态估计", 《中国电机工程学报》 *
贾小刚: "基于多渐消因子强跟踪SRUKF的故障参数辨识", 《现代电子技术》 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112114255A (en) * 2020-09-18 2020-12-22 中国计量大学 Lithium battery state of charge estimation method based on combined ensemble Kalman filtering
CN112114255B (en) * 2020-09-18 2023-02-28 中国计量大学 Lithium battery state-of-charge estimation method based on combined set Kalman filtering
CN113075561A (en) * 2021-03-23 2021-07-06 广东技术师范大学 Method, system, equipment and medium for dynamically estimating SOC of lithium ion power battery
CN113805075A (en) * 2021-09-15 2021-12-17 上海电机学院 BCRLS-UKF-based lithium battery state of charge estimation method
CN116203432A (en) * 2023-03-23 2023-06-02 广东工业大学 CSO optimization-based unscented Kalman filtering method for predicting battery state of charge
CN116203432B (en) * 2023-03-23 2023-10-20 广东工业大学 CSO optimization-based unscented Kalman filtering method for predicting battery state of charge

Also Published As

Publication number Publication date
CN111650517B (en) 2021-08-17

Similar Documents

Publication Publication Date Title
CN111650517B (en) Battery state of charge estimation method
Xu et al. State of charge estimation for lithium-ion batteries based on adaptive dual Kalman filter
CN110398691B (en) Lithium ion power battery SoC estimation method based on improved self-adaptive double unscented Kalman filter
CN110441694B (en) Lithium battery state-of-charge estimation method based on multiple fading factors Kalman filtering
Wang et al. An adaptive working state iterative calculation method of the power battery by using the improved Kalman filtering algorithm and considering the relaxation effect
CN110632528B (en) Lithium battery SOH estimation method based on internal resistance detection
He et al. State-of-charge estimation of lithium ion batteries based on adaptive iterative extended Kalman filter
CN109633479B (en) Lithium battery SOC online estimation method based on embedded type volume Kalman filtering
CN109839596B (en) SOC estimation method based on UD decomposition and adaptive extended Kalman filtering
CN109633473B (en) Distributed battery pack state of charge estimation algorithm
CN114740357A (en) Joint estimation method for branch current, charge state and power state of parallel battery pack
CN112630659A (en) Lithium battery SOC estimation method based on improved BP-EKF algorithm
CN112083333A (en) Power battery pack state of charge estimation method based on machine learning model
CN112269133A (en) SOC estimation method based on pre-charging circuit model parameter identification
CN115407205A (en) Multi-time scale SOC and SOH collaborative estimation method considering temperature influence
Manthopoulos et al. A review and comparison of lithium-ion battery SOC estimation methods for electric vehicles
CN113671377A (en) Lithium battery SOC estimation method based on improved EKF algorithm
Jin et al. Improved algorithm based on AEKF for state of charge estimation of lithium-ion battery
CN117420447A (en) Lithium battery SOC estimation method and system considering noise deviation compensation and electronic device
CN116718920A (en) Lithium battery SOC estimation method based on RNN (RNN-based optimized extended Kalman filter)
CN112485675A (en) Lithium battery SOC estimation method and system
CN116794517A (en) Lithium ion battery SOC estimation method and system based on fractional order Kalman filtering
Hu et al. Lithium battery soc estimation based on improved unscented kalman filter
CN115656848A (en) Lithium battery SOC estimation method based on capacity correction
CN115792625A (en) Lithium battery SOC estimation method, system, equipment and medium based on neural network

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