The present invention relates to a device for ascertaining state variables and/or parameters of a mathematical energy accumulator model, in particular a battery model and a corresponding method.
In electrical networks such as vehicle electrical systems that receive electrical power from a battery, it is important, in particular from the standpoint of safety aspects, to know the prevailing battery performance. In the area of automotive engineering, it is known that the state of charge (SOC) or the state of health (SOH) of the battery may be calculated from the prevailing operating parameters of the battery, in particular the battery voltage, the battery current and the battery temperature. This type of battery state detection provides information about the state of charge and/or the state of health of the battery, but it does not provide any information about the power actually deliverable or the withdrawable charge of the battery. For this reason, mathematical models of energy accumulators have already been proposed so that the chargeability, i.e., the withdrawable charge of an electrical energy accumulator, may be determined with greater precision. These are mathematical models that represent the electrical properties of the energy accumulator on the basis of various mathematical equations. These mathematical equations describe functional relationships among various state variables, e.g., battery internal voltages, currents, etc., and include various parameters that map the electrical and, if applicable, also thermal properties of the energy accumulator.
Known battery models usually include a state variable and parameter estimator, stored as software in a control unit with which the various state variables and parameters of the energy accumulator model are calculated, and a predictor capable of making a prediction about a future battery state (e.g., a battery voltage) or about the charge withdrawable from the battery, taking into account a predetermined charge profile, e.g., a current characteristic or a charge characteristic. However, known state variable and parameter estimators yield relatively inaccurate results for the desired parameters, in particular for the capacitance or the internal resistance of the battery, and therefore are not suitable for a model-based prediction or have a relatively complex structure.
The object of the present invention is therefore to create a state variable and parameter estimator for electrical energy accumulators that will map the operating performance of the electrical energy accumulator with greater precision and will thus permit better prediction of battery variables or parameters.
An important aspect of the present invention is to calculate a battery performance quantity (e.g., the battery current or the terminal voltage) with the help of a mathematical battery model with which a proportional, an integral, and a differential component of an error are determined between a measured battery performance quantity (battery voltage, current or temperature) and the battery performance quantity calculated by the mathematical energy accumulator model and to weight the individual error components each with an individual weighting factor and to couple the weighted components back into the energy accumulator model for correction purposes. A separate weighting factor is preferably assigned to each state variable and each parameter. The energy accumulator model includes various correction equations by which the state variables and/or parameters of the battery model may be corrected, taking into account the weighted error components to adjust them to the actual behavior of the energy accumulator. On the basis of the state variables and parameters corrected in this way, various performance quantities, e.g., the charge state or the withdrawable charge from the battery, may then be calculated or various battery variables may be predicted by model-based prediction.
Such a state variable and parameter estimator having feedback of a proportional, integral, and differentiated and weighed error component has the important advantage that dynamic errors may be compensated much better in particular by the differentiated component as well as static (offset) errors being compensatable in particular by the integral component. The additional feedback and taking into account the differentiated error component as well as the integral error component results in a much greater correction effect than, for example, exclusive feedback of the proportional component, i.e., the absolute error. The state variables and parameters of the battery model may therefore be determined more accurately and less ambiguously and the future battery state may be predicted with greater accuracy than when using a purely proportional error feedback.
In the case of rapid changes in charge, for example, the ohmic internal resistance may be ascertained with greater precision due to the feedback of the weighted error differential because in this way dynamic errors make a greater contribution toward the correction of the internal resistance. In a comparable manner, errors in state variables or parameters that cause a static offset error between modeled and measured performance quantities (e.g., a deviation in the initial charge state from the real value) are compensated in particular by the integral error component.
The individual feedback error components are preferably weighted as a function of an operating point. The operating point is defined in general by the prevailing values of the input variables (battery voltage, current or temperature) and the state variables of the battery model. For example, weighting of the feedback error components as a function of an operating point makes it possible to take into account only the differentiated error component for correction of the “internal resistance” state variable during the charging operation of the battery and to take into account all three error components during discharge operation. This is advisable in the case of correction of the internal resistance because in the charging case the usual battery models simulate the absolute characteristic of the battery performance quantities only inaccurately in contrast to the relative changes therein.
The weighting of the feedback error components as a function of the operating point is preferably implemented in such a way that the state variable and/or the parameter, which at the respective operating point has the greatest effect on the time characteristic, the integrated characteristic or the differentiated characteristic of the battery performance quantity to be compensated, is corrected to the greatest extent. The error between the measured battery variable and the model-based calculated performance quantity may therefore be compensated rapidly.
In addition, the integrated error component is preferably not taken into account for correction of a state variable or a parameter when their correction equations already have an integral characteristic. For example, in correction of the open-circuit voltage or the internal resistance, the integrated error component is preferably not taken into account.
The state variable and parameter estimator according to the present invention preferably includes a single energy accumulator model which is valid for the entire working range of the energy storage device. Use of different partial models for different operating ranges is not necessary in this case, so much that less effort is required to create the energy accumulator model.
Battery model 3 is a mathematical model that represents the electrical or thermal battery properties, based on various physical effects, on the basis of various mathematical models. The mathematical models describe functional correlations between the state variables, e.g., voltages, currents, temperature, etc., and include various parameters p.
Battery model 3 is implemented in the present example as a universal model that describes the behavior of battery 1 in the entire operating range.
Battery model 3 processes one or more input variables u and uses them to calculate state variables x and parameters p describing the prevailing state of battery 1. The term “input variable” is understood to include one or at most two external measured variables, e.g., “battery current” IBatt,meas, “battery temperature” TBatt,meas and “battery voltage” UBatt,meas.
In addition, battery model 3 generates an output quantity ycorr which is subtracted at node 4 from a measured battery variable ymeas of battery 1. Output quantity ycorr is an estimated battery performance quantity (current IBatt,meas, temperature TBatt,meas, voltage UBatt,meas) which is also available as a measured quantity but without the performance quantity used as input variable u.
Error dy (calculated at node 4) between model-based calculated performance quantity ycorr and measured performance quantity ymeas (e.g., battery current IBatt,meas, battery voltage UBatt,meas or battery temperature TBatt,meas) is then processed further individually for each state variable x and each parameter p and fed back into battery model 3. In the feedback path of the battery model, an integrator 8 and a differentiator 9 are provided and respectively integrate and differentiate absolute error dy. Furthermore, weighting units 5 through 7, 10, 11 are provided and individually weight the individual error components (absolute error 14, integrated error 15 and differentiated error 16) for each state variable x and each parameter p. Individual state variables x and parameters p are corrected in battery model 3 by correction equations (xcorr,k+1= . . . , pcorr,k+1= . . . ) until error dy is compensated.
The behavior of the battery may be represented in general as a nonlinear time-discrete differential equation system of the first order for implementation in a digital computer, wherein:
xk=0x0 (1)
xk+1=f(xk,p,uk) (2)
yk=h(xk,p,uk) (3)
k=0,1, . . .
where
uk=the vector of the input variables of energy accumulator model 3 in time increment k (at least one, maximum of two measured battery performance quantities IBatt,meas, UBatt,meas, TBatt,meas);
xk=the vector of state variables of energy accumulator model 3 in time increment k, e.g., open-circuit voltage UC0, polarization UD, concentration polarization Uk using starting value x0;
yk=the vector of the output quantities of energy accumulator model 3 in time increment k and includes modeled performance quantities y that are also available as measurement ymeas but without the performance quantities used as input variables uk;
p=the vector of the parameters to be estimated of energy storage device model 3;
f(xk,p,uk)=the right side of state differential equations of energy accumulator model 3 and
h(xk,p,uk)=the vector of the output equations of energy accumulator model 3.
Battery model 3 operates essentially as follows: state variables x and parameters p are first initialized at an initial state x0 and p0, respectively, and are corrected in the case of an error dy between estimated and measured output quantity y. To this end, the battery model includes correction equations (xcorr,k+1= . . . , pcorr,k+1= . . . ) which calculate corrected state variables xcorr, corrected parameters pcorr, and corrected output quantities ycorr taking into account proportional error component 14, integrated error component 15, and differentiated error component 16. Individual error components 14 through 16 are generated by an integrator 8 in the feedback path and a differentiator 9 and then weighted individually as a function of the operating point. Each weighting factor is labeled with letter K, where index p denotes weighting factor Kp for proportional error component 14, index I denotes weighting factor KI for integrated error component 15, and index D denotes weighting factor KD for differentiated error component 16. Weighted error components 17 through 19 are then added at node 12 and fed back into battery model 3 for correction of state variables x.
For correction of parameters p, weighting units 10, 11 are also provided for weighting proportional error component 14 and differentiated error component 16. Integrated error component 15 is not taken into account in the correction of parameters p because the correction equations of the parameters, e.g., internal resistance Ri or open-circuit voltage UC0, already have an integral characteristic.
Weighted error components 20, 21 are in turn added at a node 13 and fed back into battery model 3.
For the correction of state variables xcorr, parameters pcorr, and output quantities ycorr, the following equations apply:
Initial values: xcorr,0=x0,pcorr,0=p0 (4)
xcorr=x+Kp,x*(ymeas−ycorr)+K1,x*integ(ymeas−ycorr)dt+ . . . KD,x*d(ymeas−ycorr)/dt (5)
por=p+Kp,p*(ymeas−ycorr)+K1,p*integ(ymeas−ycorr)dt+ . . . KD,p*d(ymeas−ycorr)/dt (6)
ycorr=h(xcorr,pcorr,u) (7)
If x in equation (5) is replaced by f(xk,p,uk) according to equation (2) and state variables x and parameters p in equations (5), (6) are replaced by corrected quantities xcorr, pcorr, in each time increment k, the following recursive formulation of the estimation algorithm results.
xcorr,k=0=x0,pcorr,k=0=p0 (8)
xcorr,k+1=f(xcorr,k,pcorr,k,Uk)+Kp,x*(ymeas,k−ycorr,k)+ . . . K1,x*total(ymeas,1−ycorr,i)*tsample+ . . . KD,x*[(ymeas,k−ycorr,k)−(ymeas,k−1−ycorr,k−1)]/tsample (9)
pcorr,k+1=pcorr,k+Kp,p*(ymeas,k−Ycorr,k)+ . . . KD,p*(ymeas,k−ycorr,k)−(ymeas,k−1−ycorr,k−1−ycorr,k−1)]/tsample (10)
ycorr,k=h(xcorr,k,pcorrcor,k,uk) (11)
k=0,1, . . .
where tsample=sampling time.
The integral component in the parameter correction in equation (10) may be omitted because replacing the model parameters with the corrected quantities yields a feedback structure having an integral behavior. The integral component may also be omitted for state variables which already have an integrating behavior, e.g., open-circuit voltage UC0.
This structure corresponds to a state observer having additional feedback of the integrated and differentiated output error. These equations may be converted to the standard structure of a state observer having gain matrices Kx and Kp:
Initial values: xcorr,0=x0,pcorr,0=p0 (12)
xcorr,k+1=f(xcorr,k,pcorr,k,uk)+Kx*(y*meas,k−y*corr,k) (13)
pcorr,K+1=pcorr,K+Kp*(y*meas,k−y*corr,k) (14)
Kx=[Kp,xKI,xKD,x],Kp=[Kp,p0KD,p] (15)
y*corr,k=[ycorr,k;total(ycorr,1)*tsample;(ycorr,k−ycorr,k−1)/tsample] (16)
ymeas,k=[ymeas,k;total(ymeas,1)*tsample;(ymeas,k−yymeas,k−1)/tsample] (17)
k=0,1, . . .
where tsample=sampling time.
Gain matrices Kx and Kp of the error feedback may be determined by specifying the pole according to a Luenberger observer or by minimizing a quality criterion, e.g., the minimal estimated error variance (Kalman filter). In the case of nonlinear model, the model equations must first be linearized about the prevailing operating point.
Performance Quantities:
IBatt=battery current
UBatt=battery voltage
TBatt=battery temperature
State Variables:
UC0=open-circuit voltage
UK=concentration polarization
UDp=average polarization of the positive electrode
UDn=average polarization of the negative electrode
Parameters:
The individual quantities are attributable to different physical effects of the battery which are sufficiently well known from the literature and therefore will not be explained further.
Input variables of battery model 3 in this case include measured battery current IBatt,meas and battery temperature TBatt,meas. The output quantity of battery model 3 is battery voltage UBatt. Measurement error dUBatt is calculated from estimated battery voltage UBatt and measured battery voltage UBatt,meas.
For correction of state variables x, measured error dUBatt is in turn integrated (unit 8) and differentiated (unit 9) and individual components 14, 15, 16 are weighted individually. Weighted measurement error, Kx*(y*meas−y*corr) is then fed back into battery model 3 for correction of state variables x. State variables x and ohmic internal resistance Ri are balanced so that individual error components 14 through 16 are minimized.
Integrated measurement error dUBatt is not taken into account in the correction of internal resistance Ri and open-circuit voltage UC0 because its correction equations already have integral characteristics. In addition, proportional and integrated measurement errors 14, 15 are preferably fed back only for the discharge case and are not taken into account in the charging case to avoid faulty adaptation in charging operation due to the greater inaccuracy of the model. Ohmic internal resistance Ri is thus corrected during discharging by the proportional error as well as the differentiated error but only by differentiated error 16 during charging.
Number | Date | Country | Kind |
---|---|---|---|
10 2004 035 858 | Jul 2004 | DE | national |
Filing Document | Filing Date | Country | Kind | 371c Date |
---|---|---|---|---|
PCT/EP2005/052468 | 5/31/2005 | WO | 00 | 4/4/2008 |
Publishing Document | Publishing Date | Country | Kind |
---|---|---|---|
WO2006/010659 | 2/2/2006 | WO | A |
Number | Name | Date | Kind |
---|---|---|---|
4390841 | Martin et al. | Jun 1983 | A |
5349540 | Birkle et al. | Sep 1994 | A |
5489829 | Umida | Feb 1996 | A |
5596260 | Moravec et al. | Jan 1997 | A |
7109685 | Tate et al. | Sep 2006 | B2 |
20010009370 | Richter et al. | Jul 2001 | A1 |
20020130637 | Schoch | Sep 2002 | A1 |
20040032264 | Schoch | Feb 2004 | A1 |
20040162683 | Verbrugge et al. | Aug 2004 | A1 |
Number | Date | Country |
---|---|---|
0 505 333 | Sep 1992 | EP |
0505333 | Dec 1998 | EP |
Number | Date | Country | |
---|---|---|---|
20090048793 A1 | Feb 2009 | US |