Abstract
We derive and implement a battery control algorithm that can accommodate an arbitrary number of model parameters, with each model parameter having its own time-weighting factor, and we propose a method to determine optimal values for the time-weighting factors. Time-weighting factors are employed to give greater impact to recent data for the determination of a system’s state. We employ the (controls) methodology of weighted recursive least squares, and the time weighting corresponds to the exponential-forgetting formalism. The output from the adaptive algorithm is the battery state of charge (remaining energy), state of health (relative to the battery’s nominal performance), and predicted power capability. Results are presented for a high-power lithium ion battery.
Similar content being viewed by others
Abbreviations
- Ah :
-
Coulombic capacity, C-h/s
- A :
-
1/C D , 1/F
- b :
-
Regressed intercept
- B :
-
1/ (R ct C D ), 1/s
- C D :
-
Capacitance, F
- I :
-
Current, A
- L :
-
Number of parameters m to be regressed adaptively
- m :
-
Parameter to be regressed
- N :
-
Number of time steps (data points) in the regression
- P :
-
Power, W
- r :
-
Capacitance ratio,C D,discharge/C D,charge
- R :
-
High-frequency resistance, ohm
- R ct :
-
Effective interfacial resistance, ohm
- SOC :
-
Percent state of charge (energy content in the battery relative to the energy content upon full charge)
- SOH :
-
State of health, Eq. 15
- s :
-
Sum
- t :
-
Time, s
- V :
-
System voltage, V
- V H :
-
Hysteresis voltage, V
- V oc :
-
Open-circuit voltage, V
- x :
-
Time-dependent values multiplying onto parameters m
- y :
-
Dependent variable
- w SOC :
-
Weighting factor (Eq. 17)
- β:
-
Hysteresis parameter, C−1
- ɛ:
-
Error or loss term
- \({\varvec{\varepsilon}_{opt}}\) :
-
Unweighted total error as defined by Eq. 13
- λ:
-
Forgetting factor (Eq. 2)
- γ:
-
Parameter for selective weighting of data beyond that of the forgetting factor
- η I :
-
Current efficiency
- σ:
-
Variance
References
Pillar S, Perrin M, Jossen A (2001) J Power Sources 96:113
Verbrugge MW, Tate ED (2004) J Power Sources 126:236
Verbrugge MW, Liu P, Soukiazian S (2005) J Power Sources 141:369
Verbrugge MW, Frisch D, Koch B (2005) J Electrochem Soc 152:A333
Verbrugge MW, Koch BJ (2006) J Electrochem Soc 153:A187
Gelb A (ed) (1974) Applied optimal estimation. MIT Press, Cambridge, MA
Anderson BDO, Moore JB (1979) Optimal filtering. Prentice-Hall, Englewood Cliffs, NJ
Maybeck PS (1979) Stochastic models, estimation and control, vol. 141–1 of Mathematics in science and engineering, Academic Press
Widrow B, Stearn SD (1985) Adaptive signal processing. Prentice-Hall, Englewood Cliffs, NJ
Brogan WL (1985) Modern control theory, 2nd edn. Prentice-Hall, Englewood Cliffs NJ
Ljnug L, Söderström T (1986) Theory and practice of recursive identification, MIT Press
Bellanger MG (1987) Adaptive digital filters and signal analysis. Marcel Dekker, New York, NY
Åström KJ, Wittenmark B (1989) Adaptive control, Addison-Wesley
Kulhavý R (1996) Recursive nonlinear estimation. A geometric approach. Springer, London
Fortescue TR, Kershenbaum LS, Ydstie BE (1981) Automatica 17:831
Bittanti S, Bolzern P, Campi M, Coletti E (1988) Proceedings of the American Control Conference, IEEE, Austin, Texas, pp 1530–1531, December 1988
Ljung L, Gunnarsson S (1990) Automatica 26:7
Parkum JE, Poulsen NK, Holst J (1992) Int J Control 55:109
Kulhavý R (1993) Int J Control 58:905
Vahidi A, Druzhinina M, Stefanopoulou A, Peng H (2003) Proceedings of the American Control Conference, IEEE, Denver, Colorado, pp 4951–4956, June 2003
Zheng Y, Lin Z (2003) IEEE transactions on circuits and systems—II: analogue and digital signal processing 50:602
Berndt D (1993) Maintenance-Free Batteries. Research Studies Press, Taunton, Somerset, UK
Conte SD, de Boor C (1980) Elementary numerical analysis, 3rd edn. McGraw-Hill, New York, NY
Newman J, Tiedemann W (1975) AIChE J 21:25
Newman J (1991) Electrochemical systems, 2nd edn. Prentice-Hall, Englewood Cliffs, NJ
Verbrugge MW (1995) J Electrostatics 34:61
Verbrugge MW (1995) AIChE J 41:1550
Baker DR, Verbrugge MW (1999) J Electrochem Soc 146:2413
Thomas KE, Darling RM, Newman J (2002) In: van Schalkwaijk W, Scrosati B (eds) chap. 12, Advances in lithium-ion batteries. Kluwer Academic/Plenum
Beyer WH (ed) (1976) Standard mathematical tables, 24th edn. CRC Press, Cleveland OH
Acknowledgements
The authors recognize and thank Ramona Ying of General Motors R&D (Chemical and Environmental Sciences Laboratory) for acquiring the lithium ion battery data and helpful discussions with Damon Frisch and Brian Koch of the GM Electrical Center as well as Mutasim Salman of the Electrical Controls and Integration Laboratory at GM R&D.
Author information
Authors and Affiliations
Corresponding author
Appendix
Appendix
We provide a brief recapitulation of the model described in Reference [5] that is appropriate for the purposes of this work. The state of charge is taken as a weighted average (weight factor w SOC ) of values extracted by coulomb integration and voltage-based modeling:
For the coulomb-based state of charge, SOC C ,
The voltage-based state of charge, SOC V , can be determined by inverting a voltage expression for the cell derived from the equivalent circuit depicted in Fig. 1 so as to extract the open-circuit potential,
These two equations can be recast in recursive forms as
and
Time is represented by t and I denotes current; discharge currents are taken as negative. The nominal capacity Ah nominal corresponds to the ampere-hours of capacity the battery delivers when discharged from 100% SOC to 0% SOC at low rates of discharge. The self-discharge rate S D and the current efficiency η I are expected to vary with both temperature and SOC. The factor 3,600 has units of s/h, and the factor 100 is employed to keep a consistent percent basis. The parameters A and B correspond to A = 1/C D and B = 1/(R ct C D ) = 1/τ, where τ can be viewed as a time constant. A d is the inverse of the capacitance on discharge, and r is the ratio of A for charge to that of discharge; i.e.,
The open-circuit potential V oc is a function of temperature, SOC V , and a hysteresis function:
A look-up table can be used to determine the SOC V once the value of V o is obtained. For the hysteresis contribution, we construct the following first-order differential equation to calculate a hysteresis voltage V H :
or
For prolonged charge currents, or short but very large charge currents, the hysteresis voltage tends to about V H,max. The exact opposite holds for discharge currents, in which case the hysteresis voltage tends to −V H,max. Note also that if the current remains at zero for a long time, the hysteresis voltage tends to the charge-decreasing condition through self-discharge. The parameters in this equation (including V H,max) can be temperature and SOC dependent. While hysteresis plays a critical role in NiMH batteries, it is far less important in lead acid and lithium ion systems.
By combining the hysteresis and cell voltage expressions, we obtain
where E = exp(−Δt/τ). Equation 26 is the basis for the assignments provided in Eq. 12.
We now construct the power-projection capability. First, note that the max discharge power can be expressed as:
That is, when the battery voltage obtains its lowest acceptable value, the max discharge power results. We shall refer to the ohmic battery power capability as
consistent with V = V oc + IR for an ohmic battery. Similarly, the max charge power of the ohmic battery is given by
For the maximum ohmic resistance, obtained at long times (low frequency), R is replaced by R + R ct , where R ct is different for charge and discharge (cf. Fig. 1).
The ohmic battery does not address transient effects such as those correlated by the superposition integral. To improve the estimate, we employ Eq. 26 and calculate the maximum charge and discharge powers available for the time interval Δt:
where it is recognized that r = 1 on discharge. To implement these equations, the respective powers are calculated immediately after the algorithm has been employed to finish the SOC determination at time t. In this case, quantities calculated or measured at time t are then stored in the variables listed in the respective power expressions at time t−Δt. Then one must state the duration corresponding to the desired estimate for power. For example, if we want to know the power estimates 3 s from “now”, then the measured and extracted values are placed in the t−Δt quantities, Δt is set to 3 s, and the right sides of the above equations yield the desired power estimates.
The final topic of this Appendix is the skewness test of the current source. Following the procedure described in References [4] and [5], we restrict the skewness test to the actual current values I(t) and do not incorporate the charge-discharge weighting. The subscript s is added to indicate quantities associated with the skewness calculation:
and
For all of the skewness calculations employed in this work, λ = 0.99 for Eqs. 31. To start the recursive calculations for skewness, the following conditions are used.
For the first time step, skewness| t=0 is set to skew_cal(we employ a value of skew_cal = 10). Last, we define
The adaptive parameter regression analysis is not employed if the skew test is not passed, in which case:
In summary, the parameter vector m is not updated skew_test = 0; however, the approach allows one to calculate the open-circuit voltage, and SOC V can always be calculated through the use of previously extracted parameters.
Rights and permissions
About this article
Cite this article
Verbrugge, M. Adaptive, multi-parameter battery state estimator with optimized time-weighting factors. J Appl Electrochem 37, 605–616 (2007). https://doi.org/10.1007/s10800-007-9291-7
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10800-007-9291-7