See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/273489914

Optimal Battery Charging, Part I: Minimizing Time-to-Charge, Energy Loss,

and Temperature Rise for OCV-Resistance Battery Model

Article in Journal of Power Sources · February 2015

DOI: 10.1016/j.jpowsour.2015.02.075

CITATIONS READS

63 1,127

7 authors, including:

Xu Han Gopi Vinod Avvari

University of Connecticut University of Connecticut
22 PUBLICATIONS 235 CITATIONS 29 PUBLICATIONS 368 CITATIONS

SEE PROFILE SEE PROFILE

Niranjan Raghunathan Krishna R. Pattipati

University of Connecticut University of Connecticut
7 PUBLICATIONS 113 CITATIONS 404 PUBLICATIONS 7,905 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Naval Meteorology and Oceanography Command Project View project

Proactive Decision Support View project

All content following this page was uploaded by Yaakov bar-shalom on 21 February 2018.

The user has requested enhancement of the downloaded file.

 Journal of Power Sources 303 (2016) 388e398

Contents lists available at ScienceDirect

Journal of Power Sources

journal homepage: www.elsevier.com/locate/jpowsour

Optimal battery charging, Part I: Minimizing time-to-charge, energy

loss, and temperature rise for OCV-resistance battery model
A. Abdollahi, X. Han, G.V. Avvari, N. Raghunathan, B. Balasingam*, K.R. Pattipati,
Y. Bar-Shalom
Department of Electrical and Computer Engineering, University of Connecticut, 371 Fairfield Way, U-4157, Storrs, CT 06269, USA

h i g h l i g h t s

Closed form solution for minimizing weighted sum of time-to-charge and energy loss.

Semi-closed form solution by adding the temperature rise index to the cost function.

Approximating temperature rise effect as a constant heating equivalent resistance.

Analysis of efficiency performance of commercial batteries.

a r t i c l e i n f o a b s t r a c t

Article history: In this paper we present a closed-form solution to the problem of optimally charging a Li-ion battery. A
Received 23 May 2014 combination of three cost functions is considered as the objective function: time-to-charge (TTC), energy
Received in revised form losses (EL), and a temperature rise index (TRI). First, we consider the cost function of the optimization
9 February 2015
problem as a weighted sum of TTC and EL. We show that the optimal charging strategy in this case is the
Accepted 12 February 2015
Available online 18 February 2015
well-known Constant CurrenteConstant Voltage (CCeCV) policy with the value of the current in the CC
stage being a function of the ratio of weighting on TTC and EL and of the resistance of the battery. Then,
we extend the cost function to a weighted sum of TTC, EL and TRI and derive an analytical solution for the
Keywords:
Battery charging
problem. It is shown that the analytical solution can be approximated by a CCeCV with the value of
Optimal charging current in the CC stage being a function of ratio of weighting on TTC and EL, resistance of the battery and
Time to charge the effective thermal resistance.
Open circuit voltage © 2015 Elsevier B.V. All rights reserved.
State of charge

1. Introduction The most widely-used traditional strategy is the constant-current

constant-voltage (CCeCV) [9] strategy, in which a constant cur-
Battery charging is a problem of significant interest, especially as rent is applied to the battery until the terminal voltage reaches a
the battery-dependent smart devices proliferate. The literature specified value, and afterward the charging current decreases by
abounds with different strategies for charging batteries. Among the applying a constant voltage to the terminals of the battery. In Refs.
traditional methods of charging, the simplest is the constant trickle [21,22], a multi-step constant-current charging is devised for
current charge strategy, which, due to its low charging current, shortening the charging time and prolonging the cycle life of the
requires a long charging time (around 10 h) [9]; constant current battery. Using orthogonal arrays, Taguchi-based methods for bat-
strategy with higher rates of current requires shorter charging time. tery charging [12,23] present a systematic method to find the
optimal solution with guidelines for choosing the design parame-
ters. In Ref. [13], a boost charging strategy is proposed by applying
very high currents to close-to-fully discharged batteries. In pulse-
* Corresponding author.
E-mail addresses: [email protected] (A. Abdollahi), xuh06002@engr.
charging methods [6,7,17,20,27], the battery is exposed to very
uconn.edu (X. Han), [email protected] (G.V. Avvari), niranjan.raghunathan@ short rest or even deliberate discharging periods during the
uconn.edu (N. Raghunathan), [email protected], [email protected] charging process. Soft-computing approaches are also used in the
(B. Balasingam), [email protected] (K.R. Pattipati), [email protected] optimization of battery charging profile. In Ref. [24], the charging
(Y. Bar-Shalom).

http://dx.doi.org/10.1016/j.jpowsour.2015.02.075
0378-7753/© 2015 Elsevier B.V. All rights reserved.
 A. Abdollahi et al. / Journal of Power Sources 303 (2016) 388e398 389

problem is viewed as an optimization problem with the objective

function of maximizing the charge within 30 min using a multi-
stage constant current charging algorithm whose optimal solution
is obtained via an ant-colony approach. In Ref. [25], a universal
voltage protocol is proposed to improve charging efficiency and
cycle life by applying a charging profile depending on the state-of-
health (SOH) of the battery, using SOH estimation approaches [26]
in the optimization process. Recently, in Ref. [11], battery charging
is considered as an optimization problem with cost function of
time-to-charge and energy loss (as we do in this paper), but they
have not solved the problem analytically; rather they have pre-
sented a numerical solution to the problem. Other approaches, such Fig. 2. Equivalent electrical circuit model III of battery.
as genetic algorithm and neural network based strategies [16], data
mining [2,10], and Grey-predicted charging system [8] have also
been used for charging batteries. denoted by D (in seconds). We assume that the initial and final SOC
In this paper we look at the charging problem from a fresh are known: s½0 ¼ s0 ; s½kf  ¼ skf , where kf D is the charging time. We
perspective using optimal control theory, and our goal is to find the also assume that the maximum allowed value of the terminal
optimal current profile that minimizes a specific cost function. In charging voltage is vc , that is, v½k  vc for all k.
this sense, different objectives may be embedded in the cost The SOC dynamics for the battery considering the foregoing
function. One obvious cost function is the time-to-charge (TTC). We model are as follows:
prefer to minimize the charging time as much as possible, as TTC
s½k þ 1 ¼ s½k þ ch i½k (1)
reduction contributes to user satisfaction. Another important
objective is the energy loss (EL) during charging. Reducing the where ch (in 1/Amperes) is the parameter in Coulomb counting,
energy loss increases the charging efficiency. In this paper, first we given by
use an integrated cost function that includes both the TTC and EL.
Then, we also include the effects of temperature into account, and D
the cost function is selected as a linear combination of three
ch ¼ (2)
3600Cbatt
criteria: time-to-charge, energy loss, and temperature rise index
(TRI). In both cases, analytical solutions of the optimal charging where Cbatt (in Ah) is the battery capacity, assumed to be known.
problem are derived. Let the objective function be a combination of TTC and EL. In
The paper is organized as follows. In Section 2, we derive an other words,
analytical solution for the optimal charging current profile to
minimize TTC and EL. In Section 3, we extend this approach to the f 1
kX
~J ¼ wt Jt þ wE JE ¼ wt k D þ wE R0 i2 ½kD (3)
case where temperature rise is considered as well. Section 4 is tE f
devoted to simulation results and finally we conclude the paper in k¼0
Section 5.
where Jt is the TTC cost function, JE is the EL cost function; wt and
wE are weights on the TTC and EL cost functions, respectively. The
2. Analytic solution for optimal charging current profile resistance of the battery, i.e., R0 , is assumed to be known.
The charging problem then could be formulated as follows:
We consider a simplified equivalent circuit model of the battery Minimize ~J tE subject to:
as shown in Fig. 1. The theory extends naturally to more complex
h i
models involving parallel RC elements (shown in Fig. 2), but s½k þ 1 ¼ s½k þ ch i½k s½0 ¼ s0 s kf ¼ skf (4)
analytical closed form solutions are not possible in the latter case.
The model consists of a voltage source corresponding to the open-
circuit voltage (OCV), which is dependent on the state of charge V0 ðs½kÞ þ R0 i½k  vmax (5)
(SOC), and a resistance R0 . The SOC and OCV, are represented
respectively by s and V0 . The OCV is a nonlinear function of SOC and i½k  imax (6)
is denoted by V0 ðs½kÞ.
The state of charge is zero when the battery is totally discharged It is important to note that only the ratio of weights affects the
and it is one if it is completely charged. The sampling time is optimal current profile of i½k. Therefore, by dividing (3) by wE, we
redefine the cost function as follows:

. f 1
kX
JtE ¼ ~J tE wE ¼ rt Jt þ JE ¼ rt kf D þ R0 i2 ½kD (7)
k¼0

where rt ¼ wt =wE . Also note that when the current is injected into
the battery, the V0 starts to increase and this, in turn, causes the
terminal voltage to rise, until it reaches vmax , which is the
maximum allowed terminal voltage. During the whole charging
process the current should not exceed imax , which is the maximum
allowed charging current. In this paper, we use vc for vmax, where vc
is the voltage corresponding to SOC of 1; that is

vc ¼ V0 ð1Þ (8)
Fig. 1. Equivalent electrical circuit model I of battery.
390 A. Abdollahi et al. / Journal of Power Sources 303 (2016) 388e398

Assume that at time k1 , the terminal voltage v½k1  reaches vc and

let us denote the state of charge at time k1 as s1 . After time k1 , the vH½k
¼0 (17)
terminal voltage should be fixed at the constant voltage (CV) vc ; vi½k
hence, for k ¼ k1 ; k1 þ 1; …; kf  1, the dynamics of the system are
as follows: vH½k
l½k ¼ (18)
vs½k
1
i½k ¼ ðvc  V0 ðs½kÞÞ (9)
R0 vH½k
s½k þ 1 ¼ (19)
vl½k þ 1
s½k þ 1 ¼ s½k þ ch i½k (10)
From (17) we have
h i c l½k þ 1
s½k1  ¼ s1 s kf ¼ skf (11) i ½k ¼  h k ¼ 0; 1; …k1  1 (20)
2R0 D
Before going further, let us define a new equivalent problem as From (18), we can write
follows:
Minimize l½k ¼ l½k þ 1 k ¼ k1  1; …; 0 l½k1  ¼ n (21)

kX
1 1 where n is the Lagrange multiplier associated with the constraint
JtE ¼ rt Jt þ JE ¼ rt k1 D þ R0 i2 ½kD (12) s½k1  ¼ s1 . Equation (21) implies that all co-states are equal;
k¼0 therefore, we can write
subject to: l½k ¼ n k ¼ 0; 1; …k1 (22)

s½k þ 1 ¼ s½k þ ch i½k s½ 0  ¼ s0 s½k1  ¼ s1 (13) Based on (22), equation (20) can be written as

This problem is in fact the minimization in the stage where the c n

i ½k ¼  h k ¼ 0; 1; …; k1  1 (23)
terminal voltage is below vc and therefore here the condition 2R0 D
V0 ðs½kÞ þ R0 i½k  vc is not shown as we know that it holds.
Note that equation (23) states that the optimal current is con-
Inspired by Refs. [18] and [1], we solve the problem in three
stant. From (19), we can write
steps as described below:
s½k þ 1 ¼ s½k þ ch i½k (24)
❶ Given k1 (when the terminal voltage constraint becomes active),
find the optimal current profile that minimizes the energy los- which is actually the dynamics of the system. Knowing the initial
ses, and calculate the corresponding energy losses as a function state of charge (s0 ), and noting the optimal current in (23) is con-
of k1 . stant, we have
❷ Generate a new equivalent cost function JtE consisting of the
weighted TTC plus the k1 -dependent minimum energy loss X
k1 kc2h n
obtained in step 1, and find the optimal k1 based on this cost s½k ¼ s0 þ ch i½l ¼ s0  (25)
2R0 D
function. l¼0
❸ Given the optimal k1 from step 2, evaluate the optimal current Since for k ¼ k1 we have s½k1  ¼ s1 , therefore
obtained in step 1.
k1 c2h n
In the first step, assuming k1 is known, we find the optimal s1 ¼ s0  (26)
2R0 D
current i ½kjk1  that minimizes the energy loss. Having this optimal
current profile, we can calculate the minimum EL cost function Solving for n, we have
JE ðk1 Þ, which is a function of k1 . In the second step, we use the
partially optimized cost function JtE ¼ rt k1 D þ JE ðk1 Þ and we find 2R Dðs1  s0 Þ
n¼ 0 (27)
the optimum value for k1, say k1. In the third step, we insert the k1 c2h
optimal final time k1 into the current i ½kjk1  (obtained in step 1) to
Inserting (27) into (23), we have
find the optimal current i ½k. The final result will be JtE  ðk Þ.
1
The first stage is formulated as follows: s1  s0
Minimize i ½k ¼ k ¼ 0; 1; …; k1  1 (28)
k1 ch

kX
1 1
Inserting (28) into the cost function, the optimal cost function,
JE ðk1 Þ ¼ R0 i2 ½kD (14) given k1 is:
k¼0
kX
1 1
 2
s1  s0 R0 Dðs1  s0 Þ2
subject to JE ðk1 Þ ¼ R0 D¼ (29)
k¼0
k1 ch k1 c2h
s½k þ 1 ¼ s½k þ ch i½k s½0 ¼ s0 s½k1  ¼ s1 (15)
Now, consider step 2 and define the cost function as
The Hamiltonian function for this problem is
R0 Dðs1  s0 Þ2
JtE ¼ rt k1 D þ JE ðk1 Þ ¼ rt k1 D þ (30)
2
H½k ¼ R0 i ½kD þ l½k þ 1ðs½k þ ch i½kÞ (16) k1 c2h

The following equations must hold for the optimal solution [5]: To find the optimum k1 , the following relations should hold:
 A. Abdollahi et al. / Journal of Power Sources 303 (2016) 388e398 391

simplification of the dynamics of the system into the condensed

JtE ðk1  1Þ  JtE ðk1 Þ (31) condition of (40) will be useful in the next section where we derive
an analytical solution when the cost function includes the sum-
JtE ðk1 þ 1Þ  JtE ðk1 Þ (32) mation of temperature rises as well.
It should be noted that the practical meaning of the parameters
Inserting (30) into (31) and (32) we obtain two second-order
of optimization problem (e.g., wt and wE in (3) and rt in (12)) is to
equations in term of k1 . Solving these equations, we get k
1 and use them in an iterative design procedure to reach the desired
kþ , respectively, for relations (31) and (32).
1 performance. For example, if the maximum allowed energy loss is
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Emax and the maximum acceptable time-to-charge is TTCmax , then
2
±1 þ 1 þ 4R0 ðsr 1cs 2
0Þ
in the design procedure, wt and wE should be selected inversely
kH
t h
proportional to TTCmax and Emax , respectively; that is wt f1=TTCmax ,
1 ¼ 2
(33)
wE f1=Emax and then iterate. Or equivalently, rt should be selected
þ þ
The optimum k1 is ceil(k 
1 ) or floor(k1 ). Since k1  k1 ¼ 1, we proportional to Emax =TTCmax ; that is rt fEmax =TTCmax and then
 þ  þ
have ceil(k1 ) ¼ floor(k1 ) ¼ round(ðk1 þ k1 Þ=2). Thus, iterate on the proportionality factor. As Emax =TTCmax is approxi-
0vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1 mately, the allowable average power loss, rt should be selected
u proportional to allowable average power loss.
u1 R0 ðs1  s0 Þ2
k1 ¼ round@t þ A (34)
4 rt c2h
3. Optimal charging problem considering temperature
A more convenient way is to treat k1 in (30) as a continuous
variable and take derivative of (30) with respect to k1 as follows: In this section, we will extend the cost function to include the
battery temperature via temperature rise index (TRI, to be defined)
vJtE ðk1 Þ R Dðs1  s0 Þ2 as well as TTC and EL. To this end, we need a temperature model for
¼ rt D  0 ¼0 (35) the battery. Refs [15] and [19] describe the temperature model of
vk1 k21 c2 h
the battery as a linear system with two states, namely, Tcore and Tair ,
sffiffiffiffiffiffi and reference [14] uses the nonlinear heat transfer equation with a
s  s0 R0 single state. Simulations show that the dynamics of Tair have
k1 ¼ 1 (36) negligible fluctuations around the ambient temperature. Therefore,
ch rt
the temperature model, considered below, can be simplified to the
Note that if we neglect 1=4 in (34), the argument of the rounded linear part of the heat transfer equation
function in (34) is exactly the same as the one in (36). h i
Step 3 involves inserting (36) into (28) to find the optimum T½k þ 1 ¼ T½k  aðT½k  Tamb Þ þ bi2 k (41)
current
rffiffiffiffiffiffi where
s1  s0 rt
i ½k ¼ ¼ k ¼ 0; 1; …; k1  1 (37)
k1 ch R0 D
a¼ (42)
It is seen that the optimal current is constant and is a function of
mbatt Ch;batt REff
the weight on TTC and the series resistance. Therefore, the solution is the cooling coefficient and
of optimal time-to-charge and energy loss (OtE) problem is a
CCeCV profile with the current of the CC stage given by (37). R0 D
b¼ (43)
Following the CC stage, from k1 to kf , one has the CV stage where mbatt Ch;batt

v½k ¼ vc k ¼ k1 ; …; kf (38) Here T is the battery core temperature in kelvin (K), Tamb is the
ambient temperature in K, mbatt is the battery mass in kg, Ch;batt is
To the best of our knowledge, this is the first time that it is the heat capacity of the battery in J=ðkg$KÞ, and REff is the effective
proved that the well-known CCeCV charging profile is the optimal thermal resistance in K=W (kelvin/watt).
solution of a particular optimization problem, namely, the problem Defining temperature rise (TR) as T½k ~ ¼ T½k  T
amb and
of minimizing the weighted sum of time-to-charge and energy loss. assuming T½0 ¼ Tamb , we can write
In the sequel, this profile is referred to as OtE profile or OtE
policy. ~ þ 1 ¼ ð1  aÞT½k
T½k ~ þ bi2 ½k; ~ ¼0
T½0 (44)
Before we close this section, we point out another way of solving
the OtE problem of (12) and (13) by condensing (13) for all values of The solution of (44) is
k into a single condition. From (13) we can write
X
k1 
~ ¼b
T½k ð1  aÞk1l i2 l (45)
i½k ¼ ðs½k þ 1  s½kÞ=ch (39)
l¼0
Since (39) holds for k ¼ 0; 1; …; k1  1, and using the initial and
Equation (45) states that the temperature rise at any time is the
end values of SOC from (13) we can write:
integral of the square of current, from time zero up to that time
kX
with a “forgetting factor” of ð1  aÞ and the scaling factor b.
1 1
~ is positive for any k, the cost function including TTC, EL
Since T½k
i½l ¼ ðs1  s0 Þ=ch (40)
l¼0 and TR can be written as

Therefore, the problem of (12) and (13) is equivalent to a JtET ¼ rt Jt þ JE þ rT JT (46)

quadratic programming problem with the constraint in (40). In this
way, we are dealing with currents i½l as our unknowns. It is easy to where Jt and JE are TTC and EL as before and JT is the temperature
show that this results in the same solution as (37). This rise index (TRI) defined as follows:
392 A. Abdollahi et al. / Journal of Power Sources 303 (2016) 388e398

L with respect to l, and using (56) we find the optimal current

X
kf
  profile in the first stage as follows:
JT ¼ D T~ k (47)
k¼0 Geq ½kðs  s0 Þ
i ½k ¼  Pk 11   k ¼ 0; 1; …; k1  1 (57)
~ ¼ 0, the TRI can be written as
Since T½0 ch k¼0 1
Geq k

f 1
kX We refer to the current profile in (57) as the optimal time-to-
 
JT ¼ D T~ k þ 1 (48) charge, energy losses and temperature rise (OtET) policy. Note
k¼0 that (57) is similar to what we obtained for the OtE case. In
particular, if rT ¼ 0, then (57) will be the same as (28). Also,
Using (45) and (48), we can write (46) as follows
comparing (56) with the OtE case and noting that for
k ¼ 0; 1; …; k1  1 we can use the approximation of
f 1
kX
Req ½kzR0 ð1 þ rT REff Þ, analogous to the optimal current profile of
JtET ¼ rt kf D þ R0 i2 ½kD
(37), we can write
k¼0
(49)
f 1
kX X
k sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
kl 2 rt
þrT bD ð1  aÞ i ½lD i ½kz   k ¼ 0; 1; …; k1  1 (58)
k¼0 l¼0 R0 1 þ rT REff
which can be simplified as follows
We refer to the current profile of (58) as the near-optimal time-
0 1 to-charge, energy loss and temperature rise (NOtET) policy.
f 1
kX k1
kf X
 
JtET ¼ rt kf D þ D @R0 þ rT b ð1  aÞl Ai2 k (50)
k¼0 l¼0
4. Simulations
Simplifying the inner summation and noting that b=a ¼ R0 REff ,
we can write In this section, we present simulations based on the theoretical
foundations of the previous section.
f 1
kX
 
JtET ¼ rt kf D þ D Req ½ki2 k (51)
k¼0 4.1. Verification of the optimal solution

 
Req ½k ¼ R0 þ RT k (52) Here, we apply different levels of current and the simulation is
run until the terminal voltage reaches vc and after that a constant
  voltage of vc is applied until the battery is charged to skf . Five
RT ½k ¼ rT R0 REff 1  ð1  aÞkf k (53) different current profiles are chosen including the optimal current
profile (Fig. 3).pThe optimal current profile as mentioned before has
ffiffiffiffiffiffiffiffiffiffiffiffi
where RT ½k is the heating equivalent resistance. Assume, as before, the value of rt =R0 in the CC stage. The battery parameters of
that at time k1 , the terminal voltage v reaches its maximum Nokia BP-4L (Cell#3), given in the appendix, are used. The
allowable value of vc , and SOC reaches s1 . Given s1 and k1 , we can following simulation parameters are used: r ¼ 1, D ¼ 1ðsÞ, s0 ¼ 0,
write the cost function as skf ¼ 1.
The appendix also shows the parameters of the OCV curve
kX
1 1   (calculated based on [3]). The OCV is a function of SOC s as in Ref.
Jðs1 ; k1 Þ ¼ D Req ½ki2 k (54) [3].
k¼0
zðsÞbE þ sð1  2EÞ (59)
Note that we discarded the contributions of i½k1 ; …; i½kf  1,
because when the terminal voltage reaches vc the current is already
determined by the constrained dynamics of the system in (9); we OCVðzÞ ¼ K0 þ K1 z1 þ K2 z2 þ K3 z3 þ K4 z4 þ K5 z þ K6 lnðzÞ
also discarded the contribution of kf , i.e. rt kf D, because: firstly, k1 is þ K7 lnð1  zÞ
given; secondly, given s1 , kf  k1 is also known, which means kf is
known. An important point to note is that, while the upper bound (60)
of the summation in (54) is k1  1 , the formulation for Req, i.e., (52), and E ¼ 0:15. Fig. 3 shows the current profiles with different
considers the effect of the whole charging time and it contains kf levels of current in the CC stage. As seen from Fig. 3, at lower levels
rather than k1 . of current, the CC stage will take a longer time and the terminal
Now, given s1 and k1 , we can state the optimal charging problem voltage reaches the threshold voltage of vc at a later time. At higher
as follows: Minimize (54) subject to (40), or equivalently levels of current, however, the OCV grows more rapidly. As the
! terminal voltage is v½k ¼ V0 ðs½kÞ þ R0 i½k, at higher levels of cur-
kX
1 1
s  s0 rent the threshold voltage of vc is reached in a shorter time.
Minimize : L ¼ Jðs1 ; k1 Þ þ l i½l  1 (55)
ch Fig. 4(a) shows the cost function JtE for the five current profiles
l¼0
of Fig. 3 and Fig. 4(b) shows the corresponding current levels in the
Taking the derivative of Lagrangian L with respect to i½k for k ¼ CC stage. It is seen that the optimal current profile (i.e., profile 3)
0; 1; …; k1  1 and equating it to zero, we have: has the lowest cost function. Deviating from this profile, either by
  increasing or decreasing the current in the CC stage, results in an
l lGeq k increase in the cost function. For the lower current levels (profiles
i½k ¼  ¼ (56)
2Req ½kD 2D 1e2), the rise in the cost function is due to a rise in TTC and for
higher current levels (profiles 4e5) the rise in cost function is due
where Geq ½k ¼ 1=Req ½k is the conductance. Taking the derivative of to rise in EL.
 A. Abdollahi et al. / Journal of Power Sources 303 (2016) 388e398 393

1.5 0.105

0.1

c u rren t 0.095
1
0.09
7450 7500 7550 7600 7650
profile # 1
0.5 profile # 2
profile # 3 (optimal)
profile # 4
profile # 5
0
0 1000 2000 3000 4000 5000 6000 7000 8000
time (sec)

Fig. 3. Five different current profiles (including the optimal profile).

(a)
9580

9570
J

9560

9550
1 2 3 4 5
profile number
(b)
2
currennt

1.8

1.6

1.4
1 2 3 4 5
profile number

Fig. 4. Five different current profiles: a) cost functions b) current levels in CC stage.

4.2. Effect of weights rt result in low values of current in the CC stage. In other words, a
low rt puts less emphasis on charging time and more emphasis on
In this subsection, we use different cost functions and find the the energy losses; hence, it results in low level of current which
corresponding optimal profiles. Different values of rt from 0.1 to 0.5 provides low energy losses. On the other hand, by increasing rt ,
are chosen. Figs. 5e7, respectively, show the profiles of current, more emphasis is placed on the charging time. Consequently, the
pffiffiffiffiffi
state of charge and terminal voltage. Fig. 5 shows that low values of level of current is increased proportionally to rt to reduce the TTC.

1.4
ρ = 0.1
t
1.2 ρt = 0.25
ρt = 0.5
1
current (A)

0.8

0.6

0.4

0.2

0
0 20 40 60 80 100 120 140 160 180 200
time (min)

Fig. 5. Current profiles for different values of rt .

394 A. Abdollahi et al. / Journal of Power Sources 303 (2016) 388e398

0.9 ρ = 0.1
t
ρt = 0.25
0.8
ρt = 0.5
0.7

SOC 0.6

0.5

0.4

0.3

0.2
0 20 40 60 80 100 120 140 160 180 200
time (min)

Fig. 6. SOC profiles for different values of rt .

4.25

4.2

4.15 ρt = 0.1
ρt = 0.25
4.1
v (volt)

ρ = 0.5
t

4.05

3.95

3.9
0 20 40 60 80 100 120 140 160 180 200
time (min)

Fig. 7. Voltage profiles for different values of rt .

Fig. 6 shows the state-of-charge profiles for different values of functions of rt . As expected, high values of rt result in lower TTC.
rt . It is seen that by increasing rt , more emphasis is placed on The low TTC, however, is obtained by increasing the current level;
charging time and the SOC reaches the final value in a shorter time. as EL is proportional to the square of current, thus the high values of
Fig. 7 shows the terminal voltage profiles for different values of rt . rt result in high values of EL. The high values of EL mean that a
Note that for low values of rt , as the emphasis on energy loss is higher fraction of input power is wasted; hence it is equivalent to a
high, the corresponding current level in the CC stage is low, and decline in efficiency.
consequently, the terminal voltage reaches the threshold value of vc Fig. 9 shows the time-to-charge versus efficiency (ratio of
at a later time. Hence, the duration of the CC stage is high and the effective to total energy) curve. TTC and efficiency are two coun-
charging time is high as well. teracting objectives. For low values of rt , as less emphasis is put on
Fig. 8 shows the time-to-charge, energy losses and efficiency as TTC, the TTC is high; however, high TTC is the result of low current

200
TTC (min)

150
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

1400
1200
EL

1000
800
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Efficiency %

96
95
94
93
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
ρt

Fig. 8. TTC, EL and efficiency curves for different values of rt .

 A. Abdollahi et al. / Journal of Power Sources 303 (2016) 388e398 395

200
ρt = 0.1
η = 95.82%
190
TTC = 195 (min)

180

TTC (min)
170 ρ = 0.5
t
η = 92.87%
TTC = 148 (min)
160

150

140
92.5 93 93.5 94 94.5 95 95.5 96
Efficiency %

Fig. 9. TTC versus efficiency.

Table 1 (TRI) on optimal charging. The cost function is a weighted sum of

Battery thermal parameters. TTC (seconds), EL (Joules) and TRI (Kelvin seconds), given by
Paramter set mbatt ðkgÞ REff ðK=WÞ Ch;batt ðJ=ðkg$KÞÞ
JtET ¼ rt  TTC þ EL þ rT  TRI (61)
A 0.37824 7.8146 795
B 0.080 1.6528 168.15 We used two sets of thermal parameters, shown in Table 1.
Parameter set “A” is adopted from Ref. [19]. Parameter set “B” is a
scaled version of parameter set “A” with mbatt set as the weight of
Table 2 Nokia BP-4L. For each set of thermal parameters (“A” or “B”), the
Cost function for different schemes.
weights of the cost function are chosen as rt ¼ 1, rT ¼ 1 and rt ¼ 1,
rt rT Thermal parameters JtET rT ¼ 4. Three schemes are used: OtE (equations (37) and (38)), OtET
OtE NOtET OtET (equations (38) and (57)), and NOtET (equations (38) and (58)). The
cost function in (61) or (46) is calculated for the three schemes.
1 1 A 26,734 22,206 22,198
1 4 A 72,023 38,996 38,960
Table 2 shows the cost functions of the three schemes for different
1 1 B 14,970 14,696 14,696 weightings. As seen from this table, the cost function for the OtE
1 4 B 24,966 21,272 21,271 has the highest value. Also the difference between the cost function
of OtET and NOtET is negligible with the OtET being slightly smaller
when thermal parameter set “A” is used. For thermal parameter set
of “B”, there is visually no difference between NOtET and OtET. Due
values, which incur low energy losses and hence higher efficiency. to this negligible difference in the cost function and also since the
For example at rt ¼ 0:1, the TTC is 195 min, but the efficiency is as calculation of NOtET profile is much easier than that of the OtET, it
high as 95.82%. On the other hand, for high values of rt which place is reasonable to use NOtET rather than the OtET scheme. Also note
more emphasis on TTC, the TTC is reduced dramatically; however, that the weight on TRI results in a reduction of current, as can be
low TTC is achieved by increasing the current values, which results seen from Fig. 10. This reduction in current level results in a lower
in high energy losses and hence lower efficiency. For example, at temperature rise (see Fig. 11). In other words, energy losses with Req
rt ¼ 0:5, the TTC is as low as 148 min, but the efficiency decreases to instead of R0 can be used as a surrogate cost function for the TRI.
92.87%.
4.4. Analysis of different commercial batteries
4.3. Temperature effect
In this section, we discuss the behavior of different commercial
In this section, we consider the effect of temperature rise index batteries. The parameters of the investigated batteries are given in

OtE
1.5 OtET
NOtET
0.024
current (A)

0.022
1
0.02
222 222.5 223 223.5 224 224.5

0.5

0
0 50 100 150 200
time (min)

Fig. 10. Current profiles for rt ¼ 1; rT ¼ 1 and temperature parameter set “A”.
396 A. Abdollahi et al. / Journal of Power Sources 303 (2016) 388e398

300.5
OtE
OtET
300 298.255 NOtET

temperature (K)
298.25
299.5
298.245
222 223 224

299

298.5

298
0 50 100 150 200
time (min)

Fig. 11. Temperature profiles for rt ¼ 1; rT ¼ 1 and temperature parameter set “A”.

110

100 Sam−EB575152 (Cell−1)

Sam−EB575152 (Cell−2)
Sam−EB575152 (Cell−3)
Sam−EB575152 (Cell−4)
90 Sam−EB504465 (Cell−1)
Sam−EB504465 (Cell−2)
TTC (minute)

Sam−EB504465 (Cell−3)
Sam−EB504465 (Cell−4)
80
Sam−AB463651 (Cell−1)
Sam−AB463651 (Cell−2)
Nokia−BP−4L (Cell−1)
70 Nokia−BP−4L(Cell−2)
Nokia−BP−4L (Cell−3)
Nokia−BP−4L (Cell−4)
LG−LGIP (Cell−1)
60 LG−LGIP (Cell−2)

50
90.7 90.8 90.9 91 91.1 91.2 91.3 91.4 91.5 91.6 91.7
efficiency (%)

Fig. 12. Time to charge versus efficiency of different battery types at 25 C.

the appendix. Note that the equivalent electrical circuit parameters parameters of the models in Figs. 1 and 2, and the parameters of the
given in Table 3 are for model III (see Fig. 2). In simulations, we use OCV function in (60), were calculated using experimental data and
the summation of “R0 þ R1 ” of model III as an estimate of resistance by applying the BFG algorithms in Refs. [3,4]. These parameters are
R0 in model I. The batteries are Samsung EB575152 (four cells), listed in Tables 3 and 4 in the Appendix. Fig. 12 shows the TTC
Samsung EB504465 (four cells), Samsung AB463651 (two cells), versus efficiency for different types of batteries. Among all batte-
Nokia BP-4L (four cells), LG LGIP (two cells). ries, Sam-EB575152 (Cell 3) has the lowest efficiency (90.73%). This
Next, we apply the OtE algorithm with rt ¼ 0:5 to 16 commercial can be attributed to the high resistance of this battery, which might
batteries to investigate the times-to-charge and efficiencies of the be due to aging. Sam-EB504465 (Cell 4) has the highest TTC
batteries. The parameters of the batteries, i.e., the electrical (102 min) and Nokia BP-4L (Cell 4) has the highest efficiency. Note

5000

4500

4000

3500

3000
cost function

2500

2000
Sam−EB575152 (Cell−1)

Sam−EB575152 (Cell−2)

Sam−EB575152 (Cell−3)

Sam−EB575152 (Cell−4)

Sam−EB504465 (Cell−1)

Sam−EB504465 (Cell−2)

Sam−EB504465 (Cell−3)

Sam−EB504465 (Cell−4)

Sam−AB463651 (Cell−1)

Sam−AB463651 (Cell−2)

1500
LG−LGIP (Cell−1)

LG−LGIP (Cell−2)
Nokia (Cell−1)

Nokia (Cell−2)

Nokia (Cell−3)

Nokia (Cell−4)

1000

500

0
batteries

Fig. 13. Cost function for different battery types at 25 C, rt ¼ 0:5.

 A. Abdollahi et al. / Journal of Power Sources 303 (2016) 388e398 397

that the cells of the same battery are close to each other in terms of JE ðk1 Þ minimum energy loss cost function as a function of k1
efficiency and TTC. Considering all the cells of a battery, we can say (given k1 ), (29)
that LG-LGIP cells (circle markers) have the highest efficiency  ðk Þ
JtE optimized cost function
1
(91.4%). Fig. 13 shows the cost function values of JtE ¼ rt Jt þ JE . k time index, (1)
When TTC is weighted with weight value of rt ¼ 0:5, Sam- k1 optimal final time for constant-current charging stage,
EB575152 (Cell 2) has the best performance. (34)
kH
1 solutions to partially optimized cost function, (33)
5. Conclusion k1 time when terminal voltage reaches maximum allowed
value, (11)
The optimal charging problem involving a weighted combina- kf final value of time index, (3)
tion of time-to-charge (TTC), energy loss (EL) and temperature rise L Lagrangian function, (55)
index (TRI) was considered. The optimal TTC and EL solution (OtE) l½k co-state variable, (18)
is found to be the well-known CCeCV strategy with the value of mbatt battery mass in kg, (42)
current in the CC stage being a function of the ratio of weighting on NOtET near optimal TTC, energy loss, and temperature
TTC and EL and also the resistance of battery. To the best of our OCVðzÞ open circuit voltage, (60)
knowledge, this is the first time that it is proved that the well- OtE optimal time-to-charge and energy loss
known CCeCV charging profile is the optimal solution of a partic- OtET optimal TTC, energy loss, and temperature rise
ular optimization problem, namely, the problem of minimizing the R0 resistance in model I of battery as shown in Fig. 1, (3)
weighted sum of time-to-charge and energy loss. In addition, an REff effective thermal resistance in K/W, (53)
analytical solution for the optimal TTC, EL and TRI, referred to as Req ½k heating equivalent resistance added to the battery
OtET, was developed. Due to similarity of the structure of the OtE resistance, (52)
and OtET solutions, a near-optimal version of OtET was developed RT ½k heating equivalent resistance, (52)
(referred to as NOtET). The NOtET is a CCeCV strategy with the rt ratio of weight of TTC cost function to weight of EL cost
value of current in the CC stage being a function of the ratio of function, (12)
weighting on TTC and EL, the resistance of the battery and the s state of charge, (1)
effective thermal resistance. A number of simulations were con- s0 initial SOC, (4)
ducted to evaluate the effect of weighting parameters. Finally, s1 state of charge when terminal voltage reaches maximum
extensive results on industrial batteries from LG, Nokia and Sam- allowed value, (11)
sung were presented. skf final SOC, (4)
T battery core temperature in kelvin, (41)
Acknowledgments Tamb ambient temperature in kelvin, (41)
~
T½k temperature rise, (44)
The work reported in this paper was partially supported by NSF TTC time-to-charge
grants ECCS-0931956 (NSF CPS), ECCS-1001445 (NSF GOALI), CCF- v½k terminal voltage of battery, (38)
1331850 (NSF CyberSEES), ARO grant W911NF-10-1-0369, and ONR V0 open-circuit voltage, (5)
grant N00014-10-1-0029. We thank NSF, ARO and ONR for their vc voltage corresponding to SOC of 1, (8)
support of this work. Y. Bar-Shalom was partially supported by wE weight for energy loss cost function, (3)
Grant ARO W991NF-10-1-0369. Any opinions expressed in this wt weight for time to charge cost function, (3)
paper are solely those of the authors and do not represent those of zðsÞ parameter relating OCV to SOC, (59)
the sponsors.

Nomenclature Appendix

a, b coefficients in temperature model, (42,43) The following tables show the parameters of the equivalent
Cbatt capacity, (2) electrical circuit model III for different commercial batteries.
ch parameter in coulomb counting, (1)
Ch;batt heat capacity of the battery in J=ðkg$KÞ, (42)
CC constant current
CV constant voltage Table 3
Electrical parameters of model III for commercial batteries.
D sampling time, (2)
EL energy loss Make Model Cell# R0 (mU) R1 (mU) C1 (F) a Cbatt (Ah)
h charging efficiency Samsung EB575152 1 253 106 4581 0.997934 1.1875
Geq ½k conductance equal to reciprocal of Req ½k, (56) Samsung EB575152 2 209 94 5203 0.997962 1.2187
H½k Hamiltonian function, (16) Samsung EB575152 3 418 58 6222 0.99724 1.2001
i½k charging current, (1) Samsung EB575152 4 200 142 3046 0.997689 1.485
Samsung EB504465 1 259 106 4598 0.997941 1.5001
i ½k optimal current, (28) Samsung EB504465 2 268 168 2493 0.997615 1.5293
i ½kjk1  optimal current given k1 Samsung EB504465 3 272 211 1680 0.997186 1.5261
JE energy loss cost function, (3) Samsung EB504465 4 287 224 1589 0.997189 1.4831
JT temperature rise cost function, (46) Samsung AB463651 1 451 198 2100 0.997597 0.9791
Samsung AB463651 2 294 214 1950 0.997602 0.9614
Jt time to charge cost function, (3)
Nokia BP-4L 1 263 100 5031 0.998012 1.5514
JtE objective function as a combination of TTC and EL cost Nokia BP-4L 2 264 64 8141 0.99808 1.5691
functions, (12) Nokia BP-4L 3 258 95 5306 0.998028 1.5612
JtE partially optimized cost function, (30) Nokia BP-4L 4 228 50 10502 0.998106 1.613
JtET cost function including time-to-charge (TTC), energy loss LG LGIP 1 264 101 4747 0.997919 1.1141
LG LGIP 2 297 76 6654 0.998021 1.1121
(EL), and temperature rise index (TRI), (46)
398 A. Abdollahi et al. / Journal of Power Sources 303 (2016) 388e398

Table 4
OCV parameters for commercial batteries.

Make Model Cell# OCV parameters

K0 K1 K2 K3 K4 K5 K6 K7

Samsung EB575152 1 1.1288 65.0931 10.5332 1.0640 0.0457 52.2654 94.1916 0.7417
Samsung EB575152 2 1.6976 68.3306 11.1189 1.1290 0.0487 54.5366 98.4237 0.7882
Samsung EB575152 3 1.0802 68.3600 11.0617 1.1178 0.0480 55.3010 99.0817 0.8059
Samsung EB575152 4 4.3113 22.9007 4.2921 0.4926 0.0239 10.3460 27.2251 0.0226
Samsung EB504465 1 0.0218 54.9000 9.1299 0.9668 0.0442 44.6447 78.9622 0.7934
Samsung EB504465 2 1.8254 61.0951 10.0031 1.0470 0.0474 52.6013 89.7801 1.0640
Samsung EB504465 3 0.2510 55.0370 9.1256 0.9634 0.0439 45.0942 79.3945 0.8297
Samsung EB504465 4 2.9648 59.8808 9.7283 1.0127 0.0457 52.9562 88.9546 1.1311
Samsung AB463651 1 1.6972 41.8528 7.0700 0.7522 0.0343 30.6508 58.1983 0.4098
Samsung AB463651 2 1.2526 40.3216 6.7711 0.7166 0.0326 29.7536 56.3932 0.3814
Nokia BP-4L 1 3.2203 51.9246 8.8187 0.9344 0.0421 38.1050 71.7162 0.5991
Nokia BP-4L 2 2.7537 52.9707 8.9327 0.9407 0.0422 39.6357 73.7620 0.6418
Nokia BP-4L 3 3.2084 51.8554 8.7993 0.9314 0.0419 38.0572 71.6483 0.5996
Nokia BP-4L 4 2.7140 60.3626 10.0810 1.0533 0.0469 46.2542 85.0092 0.7139
LG LGIP 1 0.5267 61.5448 10.1553 1.0682 0.0485 51.2165 89.3849 0.9091
LG LGIP 2 0.4788 59.0975 9.7677 1.0290 0.0468 48.9737 85.6643 0.8748

References Vehicle Symposium, Orlando, Florida, December 15e17, 1997.

[15] A.A. Pesaran, Battery thermal models for hybrid vehicle simulations, J. Power
Sources 110 (2) (2002) 377e382.
[1] E. l. Alami, N.A. Ouansafi, N. Znaidi, On the discrete linear quadratic minimum-
[16] P. Petchjatuporn, N. Khaehintung, K. Sunat, P. Sirisuk, W. Kiranon, Imple-
time problem, J. Frankl. Inst. 335 (3) (1998) 525e532.
mentation of GA-trained GRNN for intelligent fast charger for Ni-Cd batteries,
[2] R.A. Aliev, R.R. Aliev, B. Guirimov, K. Uyar, Dynamic data mining technique for
Power Electron. Motion Control Conf. (2006) 1e5.
rules extraction in a process of battery charging, Appl. Soft Comput. 8 (3) (June
[17] B.K. Purushothaman, U. Landau, Rapid charging of lithium-ion batteries using
2008).
pulsed currents: a theoretical analysis, J. Electrochem. Soc. 153 (3) (2006).
[3] B. Balasingam, B. Pattipati, G. Avvari, K. Pattipati, Y. Bar-Shalom, A robust
[18] E.I. Verriest, F.L. Lewis, On the linear quadratic minimum-time problem, IEEE
approach to battery fuel gauging, part I: real time model identification,
Trans. Auto. Control 36 (7) (1991) 859e863.
J. Power Sources 272 (Dec. 2014) 1142e1153.
[19] J. Yan, G. Xu, H. Qian, Y. Xu, Z. Song, Model predictive control-based fast
[4] B. Balasingam, B. Pattipati, G. Avvari, K. Pattipati, Y. Bar-Shalom, A robust
charging for vehicular batteries, Energies 4 (8) (2011) 1178e1196.
approach to battery fuel gauging, part II: real time capacity estimation,
[20] J. Zhang, J. Yu, C. Cha, H. Yang, The effects of pulse charging on inner pressure
J. Power Sources 269 (Dec. 2014) 949e961.
and cycling characteristics of sealed Ni/MH batteries, J. Power Sources
[5] D.P. Bertsekas, Nonlinear Programming, Athena Scientic, Belmont, MA, 1999.
(September 2004) 180e185.
[6] L.R. Chen, A design of an optimal battery pulse charge system by frequency-
[21] T. Ikeya, N. Sawada, S. Takagi, J. Murakami, K. Kobayashi, et al., Multi-step
varied technique, IEEE Trans. Ind. Electron. 54 (1) (Feb. 2007), p. 398,405.
constant-current charging method for electric vehicle, valve-regulated, lead/
[7] L.R. Chen, Design of duty-varied voltage pulse charger for improving Li-ion
acid batteries during night time for load-leveling, J. Power Sources 75 (1) (Sep.
battery-charging response, IEEE Trans. Ind. Electron. 56 (2) (Feb. 2009).
1998) 101e107.
[8] L.R. Chen, R.C. Hsu, Chuan-Sheng Liu, A design of a grey-predicted Li-ion
[22] T. Ikeya, N. Sawada, J. Murakami, K. Kobayashi, et al., Multi-step constant-
battery charge system, IEEE Trans. Ind. Electron. 55 (10) (Oct. 2008), p. 3692,
current charging method for an electric vehicle nickel/metal hydride battery
3701.
with high-energy efficiency and long cycle life, J. Power Sources 105 (1) (Mar.
[9] R.C. Cope, Y. Podrazhansky, The art of battery charging, in: Battery Conference
2002) 6e12.
on Applications and Advances, 1999. The Fourteenth Annual, IEEE, 1999, pp.
[23] T.T. Vo, X. Chen, W. Shen, A. Kapoor, New charging strategy for lithium-ion
233e235.
batteries based on the integration of Taguchi method and state of charge
[10] G. Guo, P. Xu, Z. Bai, S. Zhou, G. Xu, B. Cao, Optimization of Ni-MH battery fast
estimation, J. Power Sources 273 (2015) 413e422.
charging in electric vehicles using dynamic data mining and ANFIS, in:
[24] Y.H. Liu, J.H. Teng, Y.C. Lin, Search for an optimal rapid charging pattern for
Advanced Intelligent Computing Theories and Applications. With Aspects of
lithium-ion batteries using ant colony system algorithm, Ind. Electron. IEEE
Artificial Intelligence, Springer, Berlin Heidelberg, 2008, pp. 468e475.
Trans. 52 (5) (2005) 1328e1336.
[11] X. Hu, S. Li, H. Peng, F. Sun, Charging time and loss optimization for LiNMC and
[25] Z. Guo, B.Y. Liaw, X. Qiu, L. Gao, C. Zhang, Optimal charging method for lithium
LiFePO4 batteries based on equivalent circuit models, J. Power Sources 239
ion batteries using a universal voltage protocol accommodating aging,
(2013) 449e457.
J. Power Sources 274 (2015) 957e964.
[12] Y.H. Liu, Y.F. Luo, Search for an optimal rapid-charging pattern for Li-ion
[26] Z. Guo, X. Qiu, G. Hou, B.Y. Liaw, C. Zhang, State of health estimation for
batteries using the Taguchi approach, IEEE Trans. Ind. Electron. 57 (12)
lithium ion batteries based on charging curves, J. Power Sources 249 (2014)
(Dec. 2010).
457e462.
[13] P.H.L. Notten, J.H.G.O. het Veld, J.R.G. Van Beek, Boost charging Li-ion batte-
[27] J. Li, E. Murphy, J. Winnick, P.A. Kohl, The effects of pulse charging on cycling
ries: a challenging new charging concept, J. Power Sources (July 2005) 89e94.
characteristics of commercial lithium-ion batteries, J. Power Sources 102 (1)
[14] A.A. Pesaran, A. Vlahinos, S.D. Burch, Thermal performance of EV and HEV
(2001) 302e309.
battery modules and packs, in: Proceedings of the 14th International Electric

View publication stats