CN108490356A - A kind of lithium battery SOC estimation method for improving EKF algorithms - Google Patents

Abstract

The present invention proposes a kind of lithium battery SOC estimating algorithms improving EKF algorithms, on the basis of EKF algorithms, thought is corrected by robust data first, with terminal voltage estimated value in EKF observational variancesWith actual measured value y_kResidual error e_kAs benchmark, using influence function, and a threshold value η is set, by comparing terminal voltage estimated valueWith actual measured value y_kResidual error and threshold value η sizes, to filtering noise variance Q_kIt is corrected in real time, reduces noise Q_kThe weight of evaluated error makes robust object function reach minimum value；Then observation noise covariance battle array Rk is adjusted by the motion interval of different SOC, influence of the model error to SOC estimation precision can be reduced.

Description

A kind of lithium battery SOC estimation method for improving EKF algorithms

Technical field

The invention belongs to technical field of lithium batteries, and in particular to a kind of lithium battery SOC estimation method for improving EKF algorithms.

Background technology

Battery is the main composition of the main energetic carrier and power resources and electric automobile whole car body of electric vehicle Battery SOC is accurately estimated in part, can not only improve the capacity utilization efficiency of battery, can also extend battery.

SOC is one of most important parameter of battery management system, accurate to estimate that only electric vehicle driver does not provide SOC Accurate remaining capacity also provides foundation for the management of battery management system and control.Currently, the most commonly used SOC estimations are calculated Method is usually Kalman filtering algorithm.Traditional Kalman filtering obtains at the standard conditions, is a kind of the linear of unbiased Minimum variance estimate algorithm, known to the mathematical model and noise statistics of dynamical system, Kalman filtering is logical It crosses measured value to be modified predictive estimation, the accurate estimation of state can be obtained.But electricity is hardly resulted in practical applications The statistical property of pond accurate mathematical model and noise, and all go estimating system state-noise to assist if filtered every time by algorithm Variance matrix Q and observation noise covariance battle array R increases the complexity of algorithm filtering.

The SOC estimation method being most widely used at present is usually Kalman filtering algorithm, is a kind of the linear of recursion Least estimated algorithm, known to system noise statistical property, estimation precision is higher.Adaptive Kalman filter is calculated Method can solve the problems, such as that nonlinear system noise statistics are unknown, all estimate noise covariance matrix due to filtering every time, Increase the complexity of algorithm filtering.

Chinese invention patent CN 104502853A disclose a kind of lithium battery SOC estimation method based on EKF, and algorithm is first Each parameter value of battery model is first calculated to finally obtain then according to the continuous iteration of Extended Kalman filter principle and data update The optimal estimation of SOC.The program can overcome in the prior art that service life is short, the poor defects such as low with reliability of safety, with Realize the advantage that service life is long, safety is good and reliability is high.But system is hardly resulted in practical applications and is accurately counted The statistical property of model and noise is learned, this may cause filtering accuracy to reduce the phenomenon that even generating filtering divergence.

Chinese invention patent CN 106291376A disclose a kind of lithium based on supporting vector machine model and Kalman filtering Battery SOC method of estimation establishes the lithium battery mould based on support vector machines by obtaining charging and discharging lithium battery historical sample data Type, then carries out recurrence calculation according to Kalman filter theory, and this method has the following advantages：Suitable for a variety of different power Battery, modeling are simple；The influence for considering external noise can effectively improve SOC estimation essences using Kalman filtering algorithm The stability of degree and prediction.But battery is a nonlinearity system, is carried out to battery SOC using Kalman filtering algorithm When estimation, nonlinearity erron is inevitably introduced.

The SOC estimation method of currently available technology is primarily present problems with：

1, model excessively simplifies or systematic parameter changes.System model in practice is typically complex, to essence True description system, the state variable in model need to reach higher dimension, this is unfavorable for that system mode is reconstructed, because This will in most cases use the method for model simplification, that is, ignore certain unessential factors in system, thereby using phase The main feature of system is described to less state variable, this may cause before application model and real system not Match.System is in actual moving process it is possible that component wear, problem of aging, these problems can cause system model It influences, and then model parameter is made to change, reduced with the matching degree of master mould.

2, noise statistics are inaccurate.It is required for considering system noise and measurement noise in a model, in most digital-to-analogue The noise statistics applied in type are all more satisfactory, cannot so as to cause the statistical property and theoretical characteristics of real process noise Reach consistency.In a practical situation, the statistical property of system may be because system interference and change that this will cause to make an uproar The inaccuracy of sound statistical property.

Invention content

The present invention propose it is a kind of based on improve EKF algorithm lithium batteries SOC evaluation method, overcome in EKF algorithms by Filtering accuracy may be caused to reduce even leads to the problem of filtering divergence in battery model error and noise statistics are unknown, it is simple The operand of adaptive Kalman filter algorithm is changed.

The invention is realized in this way：

A kind of lithium battery SOC estimation method for improving EKF algorithms, includes the following steps：

Step 1：Establish lithium battery equivalent model：

Choose the Thevenin battery models of Order RC parallel connection, the relationship of model：

U=U_ocv-R₀I-U_p1-U_p2 (1)

Wherein U is battery terminal voltage, U_ocvFor open-circuit voltage, R₁, C₁The respectively resistance and capacitance of activation polarization, R₂, C₂ The respectively resistance and capacitance of concentration polarization；

Step 2：Model parameter identification method：

Model parameter identification method reference be《FreedomCAR hybrid power automobile battery manual testings》In mention HPPC dynamic operation conditions are tested, and are tested to the SOC points of battery point different interval, the total 100s of whole process is the 1C of 10s first Pulsed discharge is then allowed to stand 40s, then carries out the pulse charge of the 1C of 10s, finally stands 40s, the SOC points of wherein HPPC experiments Respectively 0.9,0.8,0.7,0.6,0.5,0.4,0.3,0.2,0.1；Then according to Cell Experimentation An acquire data, using containing The least square recurrence method WRLS of weighted factor recognizes model parameter；

Step 3：Model parameter is recognized using the least square method of recursion containing weighted factor：

Step 3.1：Model difference equation is obtained to step 1 Chinese style (1), (2) sliding-model control：

U'(k)=k₀+k₁U'(k-1)+k₂U'(k-2)+k₃I(k)+k₄I(k-1)+k₅(k-2)

K in formula₀,k₁,k₂,k₃,k₄,k₅For undetermined coefficient.

Above formula is write as least squares formalism

In formula, ψ (k) is data vector, and θ is coefficient vector to be estimated, and k can be acquired by WRLS algorithms₀,k₁,k₂,k₃,k₄,k₅

Step 3.2：The identification of Model Parameters process of least square method of recursion based on forgetting factor：

Least square covariance P is determined first₀With the initial value of parameter matrix θ；

Then according to following recurrence formula identified parameters k₀,k₁,k₂,k₃,k₄,k₅Value, and then obtain R₀,R₁,R₂,C₁,C₂ Value；

Step 4：Battery SOC is estimated using improved EKF algorithms；

Step 4.1：Establish the discrete state-space model of battery system：

According to the mathematical equation of Thevenin battery model parameters, combine battery current integration method principle, by battery State variable of the polarizing voltage of SOC and Thevenin battery models as battery, chooses the battery terminal voltage UL conducts of measurement Observed quantity, obtained status predication equation and observation equation (3) and (4) are shown,

U_L,k=U_ocv,k-R₀i_k-Up_1,k-Up_2,k (4)

It enables

Step 4.2：Based on the estimation process for improving EKF algorithms：

1) k=0 selects initial value

2) predicted value is calculated

3) predicting covariance matrix

System noise calculates：

if|r_k|≤η

4) Kalman filtering gain matrix is calculated

Measure noise calculation：

if SOC≥SOC_H R_k=R_k0(1+G₁(SOC-SOC_H))

if SOC≤SOC_H R_k=R_k0(1+G₁(SOC-SOC_L))

Else, R_k=R_k0

5) state-updating

6) state-updating

7) judge whether filtering executes, if so, return to step 2)；Otherwise, terminate algorithm.

The advantage of the invention is that：Influence in view of battery model error and noise statistics to state estimation is Keep the state estimation of battery more accurate, the present invention proposes the estimating algorithm that a kind of collection improves EKF algorithm lithium batteries SOC, On the basis of EKF algorithms, thought is corrected by robust data first, with terminal voltage estimated value in EKF observational variancesWith reality Border measured value y_kResidual error e_kAs benchmark, using influence function, and a threshold value η is set, by comparing terminal voltage estimated value With actual measured value y_kResidual error and threshold value η sizes, to filtering noise variance Q_kIt is corrected in real time, reduces noise Q_kEstimate The weight for counting error, makes robust object function reach minimum value；Then observation noise association is adjusted by the motion interval of different SOC Variance matrix Rk, in use, battery model parameter can change battery with SOC, close close to 1 or SOC in SOC 0 section, battery variation itself is violent, and corresponding model parameter variation is also bigger, is made an uproar by the observation of the sections SOC dynamic regulation Sound variance Rk can reduce influence of the model error to SOC estimation precision.Compared with traditional EKF algorithms, which can letter The operand for changing adaptive Kalman filter algorithm, effectively overcomes the uncertainty in EKF algorithms due to battery model and makes an uproar Sound statistical property it is unknown and caused by estimation precision reduce the problem of, improve the real-time of system.

Description of the drawings

The invention will be further described in conjunction with the embodiments with reference to the accompanying drawings.

Fig. 1 is the battery model schematic diagram of the present invention.

Fig. 2 be the present invention HPPC operating modes under cell voltage change curve schematic diagram.

Fig. 3 be the present invention HPPC operating modes under battery-end curent change schematic diagram.

Fig. 4 is the measurement voltage and emulation voltage-contrast curve synoptic diagram of the present invention.

Fig. 5 is the measurement voltage and emulation voltage error curve synoptic diagram of the present invention.

Fig. 6 is the SOC algorithm estimation curve schematic diagrames of the present invention.

Fig. 7 is the SOC algorithm evaluated error schematic diagrames of the present invention.

Fig. 8 is the SOC algorithm flow charts of the present invention.

Specific implementation mode

The present invention realizes process：

A kind of lithium battery SOC estimation method for improving EKF algorithms, includes the following steps：

Step 1：Establish lithium battery equivalent model：

In view of the dynamic characteristic and complexity of battery model, what is chosen herein is the Thevenin batteries of Order RC parallel connection Model, model structure are as shown in Figure 1.

Wherein U is battery terminal voltage, U_ocvFor open-circuit voltage, R₁, C₁The respectively resistance and capacitance of activation polarization, R₂, C₂ The respectively resistance and capacitance of concentration polarization, the relationship of model：

U=U_ocv-R₀I-U_p1-U_p2 (1)

Step 2：Model parameter identification method：

Model parameter identification method reference be《FreedomCAR hybrid power automobile battery manual testings》In mention HPPC dynamic operation conditions are tested, and are tested to the SOC points of battery point different interval, the total 100s of whole process is the 1C of 10s first Pulsed discharge is then allowed to stand 40s, then carries out the pulse charge of the 1C of 10s, finally stands 40s, the SOC points of wherein HPPC experiments Respectively 0.9,0.8,0.7,0.6,0.5,0.4,0.3,0.2,0.1.HPPC test procedures such as Fig. 1 and Fig. 2 institutes of one SOC point Show, the data then acquired according to Cell Experimentation An, model is joined using the least square recurrence method WRLS containing weighted factor Number is recognized.

Step 3：Model parameter is recognized using the least square method of recursion containing weighted factor：

Step 3.1：Model difference equation is obtained to step 1 Chinese style (1), (2) sliding-model control：

U'(k)=k₀+k₁U'(k-1)+k₂U'(k-2)+k₃I(k)+k₄I(k-1)+k₅(k-2)

K in formula₀,k₁,k₂,k₃,k₄,k₅For undetermined coefficient.

Above formula is write as least squares formalism

In formula, ψ (k) is data vector, and θ is coefficient vector to be estimated, and k can be acquired by WRLS algorithms₀,k₁,k₂,k₃,k₄,k₅

Step 3.2：The identification of Model Parameters process of least square method of recursion based on forgetting factor：

Least square covariance P is determined first₀With the initial value of parameter matrix θ；

Then according to following recurrence formula identified parameters k₀,k₁,k₂,k₃,k₄,k₅Value, and then obtain R₀,R₁,R₂,C₁,C₂ Value；

Step 4：Battery SOC is estimated using improved EKF algorithms；

Step 4.1：Establish the discrete state-space model of battery system：

According to the mathematical equation of Thevenin battery model parameters, combine battery current integration method principle, by battery State variable of the polarizing voltage of SOC and Thevenin battery models as battery, chooses the battery terminal voltage UL conducts of measurement Observed quantity, obtained status predication equation and observation equation (3) and (4) are shown,

U_L,k=U_ocv,k-R₀i_k-Up_1,k-Up_2,k (4)

It enables

Step 4.2：Based on the estimation process for improving EKF algorithms：

1) k=0 selects initial value

2) predicted value is calculated

3) predicting covariance matrix

System noise calculates：

if|r_k|≤η

4) Kalman filtering gain matrix is calculated

Measure noise calculation：

if SOC≥SOC_H R_k=R_k0(1+G₁(SOC-SOC_H))

if SOC≤SOC_H R_k=R_k0(1+G₁(SOC-SOC_L))

Else, R_k=R_k0

5) state-updating

6) state-updating

7) judge whether filtering executes, if so, return to step 2)；Otherwise, terminate algorithm.

The present invention is in order to verify the accuracy of established battery model and SOC estimating algorithms, in Matlab/Simulink Establish algorithm model, and HPPC dynamic operation condition experimental datas measured according to experiment, to improve front and back SOC estimating algorithms model into Row simulation comparison is analyzed, and Fig. 4 is to emulate terminal voltage comparison diagram for HPPC dynamic operation condition drags, and Fig. 5 is error result.By Fig. 5 The experimental data as can be seen that obtained simulation result coincide substantially, maximum deviation are no more than 6%, the simulation model established It can dynamic characteristic of the good model battery in HPPC cycles.Fig. 6 is SOC estimated values and actual comparison before and after algorithm improvement Figure, Fig. 7 are error comparison diagram before and after SOC algorithm improvements, and as can be seen from Figures 6 and 7, two kinds of algorithms can accurately estimate SOC, But improved EKF algorithms evaluated error smaller is lower by 1.5% than traditional EKF algorithms estimation error no more than 4.5%.

Invention is by comparing terminal voltage estimated valueWith actual measured value y_kResidual error and threshold value η sizes, to filtering Noise variance Q_kIt is corrected in real time, and by the sections SOC dynamic regulation observation noise covariance battle array R, simplifies adaptive card The operand of Kalman Filtering algorithm improves the real-time of system.Pass through real-time update system mode noise covariance battle array Q and sight Noise covariance battle array R is surveyed, overcome in EKF algorithms makes filter due to the noise statistics design filter using mistake The problem of estimation precision reduces.

Claims (1)

1. a kind of lithium battery SOC estimation method for improving EKF algorithms, it is characterised in that：Include the following steps：

Step 1：Establish lithium battery equivalent model：

Choose the Thevenin battery models of Order RC parallel connection, the relationship of model：

U=U_ocv-R₀I-U_p1-U_p2 (1)

Wherein U is battery terminal voltage, U_ocvFor open-circuit voltage, R₁, C₁The respectively resistance and capacitance of activation polarization, R₂, C₂Respectively The resistance and capacitance of concentration polarization；

Step 2：Model parameter identification method：

The SOC points of battery point different interval are tested, the total 100s of whole process, is the 1C pulsed discharges of 10s first, then Stand 40s, then carry out the pulse charge of the 1C of 10s, finally stand 40s, the SOC points of wherein HPPC experiments are respectively 0.9,0.8, 0.7、0.6、0.5、0.4、0.3、0.2、0.1；Then the data acquired according to Cell Experimentation An, using the minimum containing weighted factor Two, which multiply recursive algorithm WRLS, recognizes model parameter；

Step 3：Model parameter is recognized using the least square method of recursion containing weighted factor：

Step 3.1：Model difference equation is obtained to step 1 Chinese style (1), (2) sliding-model control：

U'(k)=k₀+k₁U'(k-1)+k₂U'(k-2)+k₃I(k)+k₄I(k-1)+k₅(k-2)

K in formula₀,k₁,k₂,k₃,k₄,k₅For undetermined coefficient.

Above formula is write as least squares formalism

In formula, ψ (k) is data vector, and θ is coefficient vector to be estimated, and k can be acquired by WRLS algorithms₀,k₁,k₂,k₃,k₄,k₅；

Step 3.2：The identification of Model Parameters process of least square method of recursion based on forgetting factor：

Least square covariance P is determined first₀With the initial value of parameter matrix θ；

Then according to following recurrence formula identified parameters k₀,k₁,k₂,k₃,k₄,k₅Value, and then obtain R₀,R₁,R₂,C₁,C₂Value；

Step 4：Battery SOC is estimated using improved EKF algorithms：

Step 4.1：Establish the discrete state-space model of battery system：

According to the mathematical equation of Thevenin battery model parameters, combine battery current integration method principle, by the SOC of battery with State variable of the polarizing voltage of Thevenin battery models as battery chooses the battery terminal voltage UL of measurement as observation Amount, obtained status predication equation and observation equation (3) and (4) are shown,

U_L,k=U_ocv,k-R₀i_k-Up_1,k-Up_2,k (4)

It enables

Step 4.2：Based on the estimation process for improving EKF algorithms：

1) k=0 selects initial value

2) predicted value is calculated

3) predicting covariance matrix

System noise calculates：

if |r_k|≤η

4) Kalman filtering gain matrix is calculated

Measure noise calculation：

if SOC≥SOC_H R_k=R_k0(1+G₁(SOC-SOC_H))

if SOC≤SOC_H R_k=R_k0(1+G₁(SOC-SOC_L))

Else, R_k=R_k0

5) state-updating

6) state-updating

P_k|k=(I-K_kC_k)P_k|k-1

7) judge whether filtering executes, if so, return to step 2)；Otherwise, terminate algorithm.