3980 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 57, NO.

12, DECEMBER 2010

Modeling and Characterizing Supercapacitors
Using an Online Method
Nassim Rizoug, Patrick Bartholomes, and Philippe Le Moigne
AbstractIn this paper, different modeling approaches repre-
senting the behavior of supercapacitors are presented. A hybrid
modeling approach is developed to improve the characterization
of classical supercapacitor models proposed in the literature. This
approach uses various representations (frequency and temporal
results) of the same cycling test to extract easily and with a good
precision the parameters of the models. Then, it is applied for
studying the ageing of this component, which mainly depends on
the temperature and cycling process.
Index TermsAgeing, characterization, frequency response,
power electronics, supercapacitor cycling, supercapacitor model-
ing, supercapacitors, temporal response.
I. INTRODUCTION
T
HE use of supercapacitors is becoming an industrial re-
ality [1] with the improvement of manufacturing process
and the evolution of costs. Companies working in this area
have done investigation to improve this component, and its
mass production is now a real trend. Therefore, many industrial
applications are already possible [2][6]. However, one of the
main unknown is the relevance of the parameters extracted
from the measures to know the performances of this component
during its life cycle [7][11].
The aim of this paper is to understand the behavior of
supercapacitors, represent it with equivalent models, and then
use them to study the ageing of the cells (2.3 V/2700 F)
which compose the tested supercapacitor module (48 V/112 F).
Various approaches for supercapacitor modeling can be
employed:
1) electrochemical models, which are at a microscopic scale
and should be able to represent the internal phenomena
with precision;
2) equivalent circuit models, which are at a
macroscopic scale. They do not represent the
physical phenomena but are easier to manipulate because
of their reduced number of parameters.
Furthermore, circuit models of supercapacitors are very use-
ful for electrical application [12][16] because they allow the
Manuscript received April 15, 2009; revised June 18, 2009, September 9,
2009, and December 3, 2009; accepted December 17, 2009. Date of publication
February 8, 2010; date of current version November 10, 2010.
N. Rizoug is with the Ecole Suprieure des Techniques Aronautiques et
de Construction Automobile, 53061 Laval, France (e-mail: nassim.rizoug@
estaca.fr).
P. Bartholomes and P. Le Moigne are with the Ecole Centrale De Lille,
59651 Villeneuve dAscq, France (e-mail: patrick.bartholomeus@ec-lille.fr;
philippe.lemoigne@ec-lille.fr).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TIE.2010.2042418
insertion of this component in system modeling [17], [18]. The
rst necessary step is to identify the main physical phenomena
appearing inside the component and to model them with the use
of electric circuits [19]. This approach leads to distinguish be-
tween storage phenomena (capacitive effects) and energy losses
(resistance effects) appearing during the use of supercapacitors.
This paper synthesizes different approaches of modeling and
nds links between the two main well-known approaches.
The main difference between models proposed in the literature
concerns the characterization methods which are based on
frequency or on temporal responses of the component. This
paper shows the interest to choose a hybrid approach which
associates the two types of characterization (frequency and
temporal) to obtain a systematic method, with a better precision
to extract the parameters of the proposed model. This model
is then used to study the ageing of the supercapacitor module.
The characterization process is obtained using a cycling test
bench. The experimental cycle must represent the behavior of
the system for a sufciently range of frequency.
II. TEST BENCH
The behavior of models is compared to the measurements
carried out using a test bench (Figs. 1 and 2) developed in our
laboratory [20].
In this bench, two supercapacitor modules (Fig. 3) are used.
The module to characterize is composed of 24 Maxwell el-
ements (PC7223/2700 F/2.3 V) and submitted to electrically
and thermally controlled stresses. The other module (called the
buffer module) is composed of 28 elements (PC7223/2700 F/
2.3 V) and ensures the energy exchange during cycling. The
energy exchange is ensured by a reversible buck/boost con-
verter, and the losses throughout the cycling are compensated
using an auxiliary power supply.
III. MODEL AND CHARACTERIZATION
OF SUPERCAPACITORS
The transmission line approach represents the basic prin-
ciple of different supercapacitor models. These approaches are
based either on the comparison of the following:
1) the frequency response of the supercapacitors and of the
transmission line (with a frequency approach);
2) the indicial response of the supercapacitors and of the
transmission line (with a temporal approach).
Our study shows some limits on both approaches, which
depends on the chosen model topology and characterization
method. A hybrid approach is proposed to avoid drawbacks of
the previous one and to take the best of the two methods.
0278-0046/$26.00 2010 IEEE
RIZOUG et al.: MODELING AND CHARACTERIZING SUPERCAPACITORS USING AN ONLINE METHOD 3981
Fig. 1. Test bench.
A. Temporal Approach and Transmission Line
This approach of the transmission line model is based on
the representation of the charge propagation along the electrode
surface by RC circuits. Thus, the supercapacitor is considered
like a capacitance distributed along the volume, with acces-
sible areas and less accessible ones. The distribution of the
capacitance depends on the electrode material and particularly
on the geometrical properties of the pores where ions have to
access to constitute the double-layer capacitance. This approach
is well represented by an electrical circuit representing the
transmission line and using R-C parallel branches [21], [22].
Then, the proposed model will only include three parameters
(C, r
s
, and R).
According to the chosen bandwidth and its low energy
impact, the very low frequency behavior of the component is
neglected in our study. Two phenomena illustrate these slowest
effects.
The rst phenomenon concerns the redistribution of charges.
It represents the homogenization of the charges on the entire
electrode surface during the relaxation phase and can be rep-
resented by additional RC branches with long time constant
(more than 15 min).
The second phenomenon concerns the self-discharge of the
component. It can be represented by a high resistance in parallel
with the transmission line model.
The charge redistribution and the self-discharge inuence the
voltage evolution but have a negligible effect on the energetic
behavior of supercapacitors in electrical systems at the temporal
scale of power converters.
Advantages of the Temporal Approach: The temporal ap-
proach presents an interesting circuit, which is easy to associate
with the physical behavior of the supercapacitor. On the other
hand, only the data of the charge or discharge allow us to
compute the value of the capacitance and plot its evolution
according to the voltage. Thus, the capacitance can be identied
using cycles.
Drawbacks of the Temporal Approach: The identication
method of parameters (R, r
s
, and C) is based on the use of
temporal responses of the supercapacitors with charge tests us-
ing a constant current. Our tests have shown limits on methods
used for the determination of resistive parameters [23].
1) The measurement of the series resistance (r
s
) is based on
the use of the voltage drop at the end of the charge phase,
which is very difcult to measure with precision.
2) The resistance value R is quite low, and its identication
uses an approximation of a complex formula [21] which
greatly inuences the result.
On the basis of these two limits, a main drawback of this
approach concerns the identication of resistive parameters R
and r
s
.
B. Frequency Approach and Transmission Line
Several approaches have been developed by considering a
frequency model of the transmission line with a representation
using RC branches in series (Fig. 7) [24], [25]. Its frequency
response is similar to the supercapacitor one (Fig. 5). The
previous model (Fig. 4) with branches RC in parallel gives
naturally also the same frequency response [23].
The parameters of the model shown in Fig. 7 are identied
with the frequency measures (Fig. 5)
R
n
=
6(R
BF
R
HF
)
n
2

2
r
S
= R
HF
C
n
=
C
BF
2
. (1)
This classical characterization procedure is based on the
use of impedance spectrometer, injecting small currents in the
component and measuring its response.
Advantages of the Frequency Approach: The main advan-
tage of this approach concerns the determination of the resistive
parameters. These last ones are computed with a good preci-
sion by using only two points of the supercapacitor frequency
response. On the other hand, this approach is suited to the
representation of the slow phenomena, which requires long
duration (the redistribution of charges and the self-discharge).
Drawbacks of the Frequency Approach: Despite the advan-
tages of the frequency approach for resistive parameters, this
method requires an important number of tests to plot the evolu-
tion of the capacitance according to the supercapacitor voltage.
The frequency response of the supercapacitors represents the
electrical behavior of the component but is not able to give
us the link between the electrical parameters and the physical
behavior of supercapacitors. On the other hand, the use of
3982 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 57, NO. 12, DECEMBER 2010
Fig. 2. Synoptic of the test bench.
Fig. 3. Buffer module (96 F; 56 V; 400 A).
Fig. 4. Representation of the transmission line (temporal approach).
Fig. 5. Comparison between experiment and model responses (frequency
approach).
small signals may not be adapted to the representation of
the supercapacitor behavior at high currents. Finally, the fre-
quency characterization requires the manipulation of superca-
Fig. 6. Model and parameters dened with the hybrid approach.
pacitors because of the use of supplementary measuring device
(spectrometer).
C. Hybrid Approach of the Transmission Line
As explained in the previous section, the main problem of the
temporal approach concerns the low precision obtained on the
resistive parameters. At the opposite, the identication of these
resistive parameters with the frequency representation is very
interesting because of its precision and its simplicity.
Our proposition is based on the possibility of a transition
between the two approaches of the transmission line (temporal
and frequency). This link between the two approaches repre-
sents the base of the hybrid approach used in this paper. The
hybrid approach allows us to take advantages of the frequency
characterization to calculate the resistive parameters, which
improves the precision compared to temporal methods. On the
other hand, the temporal approach is preferred for the char-
acterization of the capacitance (Fig. 6) to avoid the important
number of measures needed in the frequency approach.
Fig. 6 shows the model chosen for the hybrid approach. r
s
and R
i
= R/n are computed using an online frequency charac-
terization (Fig. 5) and C(V ) using the temporal response of a
complete charge of supercapacitors (Fig. 8). The chosen circuit
(Fig. 6) has been preferred to the other one (Fig. 7) because it is
much more representative of physical phenomenon appearing
inside the component (ion mobility into porous electrode with
different accessibility).
Capacitance Characterization: The temporal response is
well adapted for the characterization of the capacitance. In this
case, the capacitance is computed using data of a charge phase.
At the opposite of the frequency identication, the capacitance
is calculated using only one charge.
Several methods can be used to evaluate different repre-
sentations of this capacitance (electrical, energetic, etc.) [23].
RIZOUG et al.: MODELING AND CHARACTERIZING SUPERCAPACITORS USING AN ONLINE METHOD 3983
Fig. 7. Frequency approach and associated transmission line circuit.
Fig. 8. Estimation of the capacitance versus the voltage.
The selected method for this paper is based on the following
relation:
I
sc
=C.dV
sc
/dt
C =aV
2
sc
+ bV
sc
+ c. (2)
In our case, a = 250.5, b = 1115, and c = 2010.
The main drawback of this method is the noise involved
by the derivation operation. The effect of this noise has
been greatly reduced by ltering and using a great number
of samples. The capacitance characteristic obtained with our
module (Fig. 8) can be represented using a polynomial approxi-
mation (2).
Online Resistance Characterization: The test bench allows
a continuous cycling of supercapacitors by applying charge
and discharge currents to the tested module. The current I
sc
in the tested module is dened with T
o
and T
ch
, as described
in Fig. 9. This gure shows a current prole corresponding to a
charge/discharge cycle
_
_
_
I
SC
= I, [T
0
/2, T
Ch
+ T
0
/2]
I
SC
= I, [T
Ch
+ 3.T
0
/2, T T
0
/2]
I
SC
= 0, Else.
(3)
This periodic cycle current can be decomposed into a Fourier
series
I
SC
= I
1
sin(2f
s
t) + I
3
sin (2(3f
s
)t) +
+I
2n+1
sin (2 ((2n + 1)f
s
) t) (4)
with
I
i
=
4I
I
cos(i
0
)
0
=
T
0
T
.
Fig. 9. Current prole in the cycle used for the characterization of
supercapacitors.
To limit the inuence of the signal noise, three criteria for the
choice of the represented harmonics have been xed.
1) Only odd harmonics are represented.
2) Only harmonics which have sufcient amplitude (higher
than 1 A) are presented.
3) The frequency range is limited to rank 100.
The decomposition in Fourier series of the temporal response
[V
sc
(t), I
sc
(t)] of the supercapacitors allows one to nd the
response of the system for each elementary frequency of the
injected signal. For each elementary frequency, one point on
the Nyquist complex plane is obtained, representing the behav-
ior of the system at this frequency.
Fig. 11 shows the experimental frequency response of the su-
percapacitor. The cycling period is about 50 s (f
s
= 20 mHz),
and the discharge duration T
ch
= Tdech = 15 s with a contin-
uous current I = 120 A (Fig. 10). Furthermore, the frequency
range has been limited to h = 100 (2 Hz) in order to have
harmonic currents higher than 1 A.
Fig. 11 shows that the charge/discharge cycles used to plot
the frequency response of the supercapacitors do not allow us to
represent the behavior of the component for the high frequency
(250 Hz). The bandwidth of the cycling signal is not sufcient
to identify precisely the series resistance (r
s
= R
HF
).
In order to expend the spectral band of the signals and to
extend the curve until the r
s
value is obtained, an additional
high-frequency current (50 Hz) of a few amperes (5 A of rms
current) is injected during cycles used for characterization.
Fig. 12 shows the waveform of the current during the data
acquisition.
Fig. 13 shows the frequency response of one element of
the supercapacitor module with the injection of the 50-Hz
3984 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 57, NO. 12, DECEMBER 2010
Fig. 10. Amplitude of the cycling harmonic current (I = 120 A).
Fig. 11. Frequency response of supercapacitors using charge/discharge
cycles.
Fig. 12. Cycling current with injection of high-frequency harmonic (f
HF
=
50 Hz).
harmonic. This response represents the behavior of the com-
ponent between 20 mHz and 50 Hz.
Validation of the Model: Figs. 14 and 15 show a comparison
between the response of the model dened with the hybrid
Fig. 13. Response of the frequency model and experimental response using
charge/discharge cycles with injection of 50-Hz harmonic.
Fig. 14. Test with complete load at 10 A.
Fig. 15. Test with complete load at 100 A.
RIZOUG et al.: MODELING AND CHARACTERIZING SUPERCAPACITORS USING AN ONLINE METHOD 3985
Fig. 16. Simplied model.
approach and the experimental response of one element of the
module for two tests of complete load (with I = 10 A and I =
100 A). The parameters of the hybrid model are the same for
the two tests [R
HF
= 0, 4478 m, R
BF
= 0, 767 m, n = 10,
and the capacitance is given by (2)]. These tests illustrate the
use of supercapacitors in two different types of application: for
the rst one, a load with low currents and low frequencies and,
for the second one, a fast load (increased frequency) with a
higher current (100 A). These results show that the response of
this model follows the experimental response, which validates
the characterization method.
Advantages and Drawbacks of the Hybrid Approach: The
hybrid approach allows us to avoid the drawbacks of the
temporal and frequency methods and use the best of the two
approaches (temporal and frequency) to obtain a good cha-
racterization precision for all the parameters (frequency
characterization for the resistive parameters and temporal
characterization for the capacitive ones).
The precision of this model is ensured for frequency higher
than about 1 mHz, corresponding to frequencies in the validity
domain of the chosen model and used in cycling tests. Below
this frequency limit, and as explained in Section III-A, the
charge redistribution at very low frequencies may become
important and have to be taken into account to obtain a good
electrical model. Then, the chosen topology (Fig. 6) has to be
modied [23].
D. Simplied Model R
eq
C
eq
Extracted From the
Transmission Line Model
The aimof this section is to precise the link between the iden-
tication method of the transmission line proposed previously
and the characterization of a simplied model R
eq
C
eq
series
(Fig. 16). This model does not take care of transmission effects
and is focused on the energetic representation. Its bandwidth
is highly reduced but sufcient for application where the trans-
mission effect has a very small duration compared to the cycle
period.
The characterization in this case is based on teh temporal
equations of the supercapacitor
_
V
SC
= R
eq
I
SC
+
I
SC
= C
eq
d
dt
.
(5)
R
eq
and C
eq
are the parameters resulting from this identica-
tion approach.
The previous transmission line model can be simplied with
an energetic approach. This last one allows the identication
Fig. 17. Current of branches 1, 5, and 10 of the transmission line model.
of the simplied model parameters (Fig. 16) using the pre-
vious characteristics r
s
, R, and C of the transmission line
(Fig. 6).
The time constant of the supercapacitors is less than 0.2 s
(Fig. 17). Thus, it is possible to consider that the currents of all
branches are the same
i
C_i
= i
C
=
I
n
. (6)
Moreover, the equivalent capacitance C
eq
is the sum off all
branch capacitances
C
eq
= n
_
C
n
_
= C (7)
where
i
c_i
current of the branch i;
I supercapacitor current;
n number of branches;
C/n capacitance of one branch of the transmission line.
Using (6), the current of the resistances (R
1
, R
2
, . . . , R
n
)
can be written as
i
R1
= I, i
R2
= I i
C
, i
R3
= I 2.i
C
, . . . i
Rn
= i
C
.
(8)
Replacing (6) in (8), we get
i
R1
=I, i
R2
=I
_
1
1
n
_
, i
R3
=I
_
1
2
n
_
, . . . , i
Rn
=
I
n
.
(9)
The equivalent resistance R
eq
of the simplied model must
represent the joule losses of all transmission line resistances (r
s
,
R1, R2, . . . , Rn)
Pj = R
eq
I
2
= r
s
.I
2
+
n

i=1
R
i
.i
2
Ri
(10)
where R
i
= R/n.
3986 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 57, NO. 12, DECEMBER 2010
Fig. 18. Variation of the rms current with a constant period cycle.
Using (9) and (10), R
eq
can be written as
R
eq
=r
s
+
R
n
_
1
2
+
_
1
1
n
_
2
+
_
1
2
n
_
2
+
_
1
3
n
_
2
+ +
_
1
n
_
2
_
=r
s
+
R
n
3
_
n
2
+ (n 1)
2
+ (n 2)
2
+(n 3)
2
+ + 1
2

. (11)
Hence
_

_
R
eq
= r
s
+
R
n
3
i=n

i=1
i
2
= r
s
+
R(2n+1)(n+1)
6.n
2
= r
s
+ R
trans
(n)
C
eq
= C.
(12)
When n tends toward innity, the formula of the equivalent
resistance R
eq
is simplied as follows:
R
eq
= r
S
+
R
3
= R
BF
. (13)
This direct link between the parameters of the two models
allows using the characterization of the transmission line to
identify the parameters of the simplied R
eq
C
eq
model.
Supercapacitor Efciency Evaluation Using the Simplied
Model: In addition to the temporal tests, the simplied model
has been validated with other experimental measures. These
tests concern the measurements of efciency and tests with
constant power. First, some trials with various values of current
were fullled using the same cycle period, but with a variation
of the charge/discharge duration, which involves a variation of
the rms cycle current (Fig. 18).
The variation of efciency according to the rms current for
three values of charge/discharge current is shown on Fig. 19.
It shows that, for the same value of the cycling current, the
efciency decreases when the rms current increases and that,
for the same rms current, there is a signicant decrease of the
efciency when the cycling current increases. The simplied
model follows these experimental tendencies which conrm
that the simplied model will be sufcient for a majority of
applications where the current spectrum contains mainly low
frequencies. The computation of the efciency using this model
gives a light overestimate which does not exceed 0.6%.
Fig. 19. Comparison between the measured and estimated efciencies of
supercapacitors.
Fig. 20. Test at constant power P = 3 kW.
Discharge Tests With Constant Power: Tests at constant
power were also made to validate the precision of the simplied
model and to check the validity of the developed models with
different cycling conditions compared to those used for the
characterization. The use of this type of test is commonly used
to dene the kind of application for storage components in
the energy power plan (Ragones diagram). Supercapacitors are
here considered to be connected to a load consuming a constant
power, and the discharge current is controlled according to the
evolution of the voltage to keep the power at constant value.
The injected current is (P/V
sc
), with V
sc
as the voltage
of the supercapacitors and P as the constant power. Fig. 20
shows the evolution of the current and the voltage of the tested
module for a constant power P = 3 kW. The initial voltage of
the module is 48 V (nominal voltage).
These tests are stopped when the voltage reaches the minimal
limit of the voltage (V min) of the test bench. Below this
voltage, the stored energy can be considered negligible. For our
test bench, V min is equal to 15 V.
To calculate the duration of the discharge at various powers,
different tests at constant power between 500 W and 6 kW
RIZOUG et al.: MODELING AND CHARACTERIZING SUPERCAPACITORS USING AN ONLINE METHOD 3987
Fig. 21. Energetic behavior of the supercapacitor module.
have been fullled. Fig. 21 shows the evolution of the discharge
duration of each test according to the xed power.
Fig. 21 shows the decrease of the discharge duration from
227 s (P = 500 W) to 15.3 s (P = 6 kW). The multiplication
of this duration for each test by the required power allows
the calculation of the energy that the module is able to re-
store. These results prove also the very good estimation of
the discharge duration using the simplied model, which is
not the case with the use of the classical manufacturer model
(C constant). The error using this last one can reach 63%.
IV. USING THE HYBRID MODEL TO VALUATE THE AGEING
OF A SUPERCAPACITOR MODULE
In order to avoid waste of time and to have ideal test
conditions, the test bench is completely automatic and does not
need any equipment handling.
The data stored along the cycling have been used to obtain
electrical parameters of the module for a few elements and,
consequently, the ageing of supercapacitors. The characteriza-
tion procedure has been regularly repeated during the cycling
process in order to evaluate the ageing of supercapacitors
(the increase of the resistive effect and the decrease of the
capacitive effect). This method based on data acquisition and
posttreatment has allowed us to evaluate the incertitude effect,
and to dene a well-adapted measurement protocol.
The choice of the cycles parameters is based on the thermal
behavior of the supercapacitor and the kind of application. In
our case, the cycles characteristics allow one to reach 80% of
energy discharge. The cycle duration is 50 s with two steps of
current of 120 A during 17 s.
These specications allow one to obtain an rms current of
99 A. Furthermore, the voltage varies between the maximum
(48 V) and the half of the maximum value (24 V). These
cycle specications allow one to maintain the maximum of the
elements temperatures under 55

C.
The cycling data stored (Fig. 22) can be exploited to test
different methods of characterization and to make several stud-
ies (characterization, efciency, energetic behavior, . . ., etc.).
These studies have been nally developed using processing
methods based on the hybrid approach previously presented.
This approach allows us to characterize the supercapacitors
module at precise moment of its lifetime. The repetition of this
characterization procedure along cycling allows the measure-
ment of the degradation of the supercapacitors characteristics
(Fig. 22).
Until now, more than 530 000 cycles have been fullled using
the presented test bench. In this part, the ageing results of one
element (element 12) are presented (Fig. 23). The evolution of
its characteristics (r
s
, R
eq
, R, and C
eq
) is compared to that of
the whole module (average characteristics).
The selected element (element 12) is submitted to a great
thermal stress as it is located at the center of the module. The
temperature of this element is xed around 46

C with the help
of the cooling system control.
A. Evolution of the Series Resistance r
s
The series resistance (r
s
) is calculated using the frequency
response of supercapacitors at 50 Hz (Fig. 5). Fig. 24 shows the
increase of series resistance with the number of cycles.
Degradation is higher for element 12 compared to the whole
module. For this element, ageing speeds up after 350 000 cycles
and reaches 32% after 530 000 cycles. This degradation does
not exceed 19% for the whole module, which proves that the
evolution of the average resistance of the module does not
represent the ageing of each component.
B. Evolution of the Equivalent Resistance R
eq
Frequency response is also used to calculate the equiva-
lent resistance R
eq
of the simplied model R
eq
C
eq
. This
resistance represents the real part of the frequency response
at the cycling frequency (20 mHz). The Fig. 25 shows the
increase of the equivalent resistance according to the number of
cycles. It speeds up after 350 000 cycles and reaches 19% after
530 000 cycles. The degradation for element 12 (the element
is more stressed) is more important compared to the average
resistance of the module (11%).
C. Evolution of the Transmission Resistance R
The use of (13) allows us to plot the evolution of the
transmission resistance R according to the number of cycles
(Fig. 26).
Fig. 26 shows the variation of the transmission resistance R
according to the number of cycles. As R decreases with
temperature [27], elements at the center of the module (as
element 12) have lower resistance compared to the average
module resistance. This parameter seems almost stable and may
not be used to estimate the ageing of the supercapacitors.
D. Evolution of the Capacitance C
In this case, the capacitance is evaluated at a xed voltage
(V
sc
= 1, 7 V), which is close to the nominal one.
Fig. 27 shows the evolution of the average capacitance of the
module and that of element 12 during the rst 530 000 cycles of
their lifetime. This gure shows the decrease of the capacitance
according to the number of cycles for the rst 440 000 cycles.
3988 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 57, NO. 12, DECEMBER 2010
Fig. 22. Continuous cycling and ageing of supercapacitors.
Fig. 23. Tested module.
Fig. 24. Evolution of r
s
according to the number of cycles.
Then, the capacitance is regenerated with the help of a 30-day
pause of the system, which produces a small increase of the
capacitance. After that, the capacitance continues to decrease;
Fig. 25. Evolution of R
eq
according to the number of cycles.
Fig. 26. Evolution of R according to the number of cycles.
this variation does not exceed 7% after 530 000 cycles. Because
of this low capacitance variation, its evolution along cycles is
very sensitive to the noise measurement.
RIZOUG et al.: MODELING AND CHARACTERIZING SUPERCAPACITORS USING AN ONLINE METHOD 3989
Fig. 27. Evolution of C according to the number of cycles.
The results presented in Fig. 27 show the same degradation
for element 12 and for the average capacitance of the module,
which proves the low inuence of temperature on the degrada-
tion of capacitance.
V. CONCLUSION
In this paper, various approaches for the modeling of super-
capacitors have been presented. These models, based on the
frequency or temporal characterization, are commonly used to
represent the behavior of the component according to some
parameters (frequency, voltage, temperature, . . .). The advan-
tages and limits of these models have been analyzed. The main
limitations concern the identication of the supercapacitors
parameters with high precision, the high number of tests, the
use of additional measuring device (impedance-meter), and the
difculty of direct characterization and diagnostic for industrial
power applications.
The proposed hybrid method allows resolving these prob-
lems. The link between the two classical models allows us
to combine the two characterization methods to improve the
precision. The frequency characterization is used for the resis-
tive parameters r
s
and R and the temporal characterization for
the capacitance C. This method, based on the use of the cy-
cling data, can be applied for industrial application diagnostics
(online method of characterization).
A simplied model using the parameters of the previous
hybrid characterization has been presented and tested in an
energetic study. It proves the possibility to use it in the case
of medium frequency, which is higher than 1 mHz.
These models are used to study the ageing of the su-
percapacitors. For that, the cycling data are regularly stored
along the lifetime of supercapacitors. Until now, more than
530 000 cycles have been made with the test bench. The results
of ageing show the increase of resistances r
s
and R
eq
according
to the number of cycles. This degradation is more important
for the more stressed element (element 12) with a decline that
reaches 32%for the series resistance and 19%for the equivalent
resistance after 530 000 cycles. The transmission resistance R is
almost stable and is not degraded. Cycling results show also the
very lowdegradation of the capacitance with a small diminution
(7%). Moreover, the capacitance degradation, in this case, does
not depend on the temperature of elements.
Finally, these tests have conrmed the high electrical and
ageing performances of supercapacitors. These performances
are improved with the help of technological and manufacturing
developments, which will place this storage component as a key
for future development of high-efciency electrical systems.
REFERENCES
[1] A. Kusko and J. DeDad, Stored energyShort-term and long-term en-
ergy storage methods, IEEE Ind. Appl. Mag., vol. 13, no. 4, pp. 66721,
Jul./Aug. 2007.
[2] C. Abbey and G. Joos, Supercapacitor energy storage for wind en-
ergy applications, IEEE Trans. Ind. Appl., vol. 43, no. 3, pp. 769776,
May/Jun. 2007.
[3] Y. Hyunjae, S. Seung-Ki, P. Yongho, and J. Jongchan, System integration
and power-ow management for a series hybrid electric vehicle using
supercapacitors and batteries, IEEE Trans. Ind. Appl., vol. 44, no. 1,
pp. 108114, Jan./Feb. 2007.
[4] P. Pillay, R. M. Samudio, M. Ahmed, and R. T. Patel, A chopper-
controlled SRM drive for reduced acoustic noise and improved ride-
through capability using supercapacitors, IEEE Trans. Ind. Appl., vol. 31,
no. 5, pp. 10291038, Sep./Oct. 1995.
[5] A. Rufer and P. Barrade, A supercapacitor-based energy-storage system
for elevators with soft commutated interface, IEEE Trans. Ind. Appl.,
vol. 38, no. 5, pp. 11511159, Sep./Oct. 2002.
[6] J. L. Duran-Gomez, P. N. Enjeti, and A. Von Jouanne, An approach to
achieve ride-through of an adjustable-speed drive with yback converter
modules powered by super capacitors, IEEE Trans. Ind. Appl., vol. 38,
no. 2, pp. 514522, Mar./Apr. 2002.
[7] D. Linzen, S. Buller, E. Karden, and R. W. De Doncker, Analysis and
evaluation of charge-balancing circuits on performance, reliability, and
lifetime of supercapacitor systems, IEEE Trans. Ind. Appl., vol. 41, no. 5,
pp. 11351141, Sep./Oct. 2005.
[8] R. A. Dougal, S. Liu, and R. E. White, Power and life extension of battery
ultracapacitor hybrids, IEEE Trans. Compon. Packag. Technol., vol. 25,
no. 1, pp. 120131, Mar. 2002.
[9] J.-N. Marie-Francoise, H. Gualous, and A. Berthon, Supercapacitor
thermal- and electrical-behaviour modelling using ANN, Proc. Inst.
Elect. Eng.Elect. Power Appl., vol. 153, no. 2, pp. 255262, Mar. 2006.
[10] R. M. Nelms, D. R. Cahela, and B. J. Tatarchuk, Using a Debye polar-
ization cell to predict double-layer capacitor performance, IEEE Trans.
Power Electron., vol. 37, no. 1, pp. 49, Jan./Feb. 2001.
[11] G. Coquery, R. Lallemand, J. Kauv, A. De Monts, B. Soucaze-Guillous,
J. Chabas, and A. Darnault, First accelerated ageing cycling test on su-
percapacitors for transportation applications: Methodology, rst results,
in Proc. ESSCAP, Belfort, France, 2004.
[12] J. Moreno, M. E. Ortuzar, and J. W. Dixon, Energy-management system
for a hybrid electric vehicle, using ultracapacitors and neural networks,
IEEE Trans. Ind. Electron., vol. 53, no. 2, pp. 614623, Apr. 2006.
[13] M. E. Ortuzar, R. E. Carmi, J. W. Dixon, and L. Moran, Voltage-source
active power lter based on multilevel converter and ultracapacitor DC
link, IEEE Trans. Ind. Electron., vol. 53, no. 2, pp. 477485, Apr. 2006.
[14] P. Thounthong, S. Rael, and B. Davat, Control strategy of fuel cell and
supercapacitors association for a distributed generation system, IEEE
Trans. Ind. Electron., vol. 54, no. 6, pp. 32253233, Dec. 2007.
[15] U. K. Madawala, D. J. Thrimawithana, and N. Kularatna, An ICPT-
supercapacitor hybrid system for surge-free power transfer, IEEE Trans.
Ind. Electron., vol. 54, no. 6, pp. 32873297, Dec. 2007.
[16] M. Ortuzar, J. Moreno, and J. Dixon, Ultracapacitor-based auxiliary
energy system for an electric vehicle: Implementation and evaluation,
IEEE Trans. Ind. Electron., vol. 54, no. 4, pp. 21472156, Aug. 2007.
[17] Y. Diab, P. Venet, H. Gualous, and G. Rojat, Self-discharge characteriza-
tion and modeling of electrochemical capacitor used for power electronics
applications, IEEE Trans. Power Electron., vol. 24, no. 2, pp. 510517,
Feb. 2009.
[18] W. Lajnef, O. Briat, S. Azzopardi, E. Woirgard, and J.-M. Vinassa, Ultra-
capacitor electrical modeling using temperature dependent parameters,
in Proc. ESSCAP, Belfort, France, 2004.
[19] N. M. L. Tan, S. Inoue, A. Kobayashi, and H. Akagi, Voltage balancing
of a 320-V, 12-F electric double-layer capacitor bank combined with a 10-
kW bidirectional isolated dcdc converter, IEEE Trans. Power Electron.,
vol. 23, no. 6, pp. 27552765, Nov. 2008.
[20] P. Bartholomes, B. Vulturescu, X. Pierre, N. Rizoug, and P. Le Moigne,
A 60 V400 A test bench for supercapacitor modules, in Proc. EPE,
Toulouse, France, Sep. 2003.
3990 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 57, NO. 12, DECEMBER 2010
[21] F. Belhachemi, S. Ral, and B. Davat, A physical based model of power
electric double-layer supercapacitors, in Conf. Rec. IEEE-IAS Annu.
Meeting, 2000, pp. 30693075.
[22] L. Zubieta and R. Bonert, Characterization of double-layer capacitors
for power electronics applications, IEEE Trans. Ind. Appl., vol. 36, no. 1,
pp. 199205, Jan./Feb. 2000.
[23] N. Rizoug, Modlisation lectrique et nergtique des supercondensa-
teurs et mthodes de caractrisation: Application au cyclage dun module
de supercondensateurs basse tension en grande puissance, M.S. thesis,
Ecole centrale de Lille, Lille, France, Feb. 2006.
[24] S. Buller, E. Karden, D. Kok, and R. W. De Doncker, Modeling the dy-
namic behavior of supercapacitors using impedance spectroscopy, IEEE
Trans. Ind. Appl., vol. 38, no. 6, pp. 16221626, Nov./Dec. 2002.
[25] S. Buller, M. Thele, R. W. A. A. De Doncker, and E. Karden, Impedance-
based simulation models of supercapacitors and Li-ion batteries for power
electronic applications, IEEE Trans. Ind. Appl., vol. 41, no. 3, pp. 742
747, May/Jun. 2005.
[26] N. Rizoug, P. Bartholomes, P. Le Moigne, and B. Vulturescu, Electrical
and thermal behaviour of a supercapacitor module: On-line characteriza-
tion, in Proc. ESSCAP, Belfort, France, 2004.
[27] M. Nobuyoshi and I. Takayoshi, A control method to charge series-
connected ultraelectric double-layer capacitors suitable for photovoltaic
generation systems combining MPPT control method, IEEE Trans. Ind.
Electron., vol. 54, no. 1, pp. 374383, Feb. 2007.
Nassim Rizoug received the Engineering degree
from the Ecole Nationale Polytechnique dAlger,
Alger, Algeria, in 1998 and the Ph.D. degree in
electrical engineering from the Ecole Centrale de
Lille, Villeneuve dAscq, France, in 2006.
He is currently an Assistant Professor with the
Ecole Suprieure des Techniques Aronautiques et
de Construction Automobile, Laval, France, since
2007. His research interest includes specially the
eld of energy management and the characterization
of the storage component for the power electronic
applications.
Patrick Bartholomes received the B.Eng. and
Ph.D. degrees from the Universit des Sciences et
Technologies de Lille, Villeneuve dAscq, France, in
1991 and 1996, respectively.
Since 1998, he has been an Assistant Professor
with the Ecole Centrale de Lille, Villeneuve dAscq.
His research interests are power electronic converter
structures and supercapacitor characterization.
Philippe Le Moigne received the Engineering de-
gree from the Institut Industriel du Nord (currently
Ecole Centrale de Lille), Villeneuve dAscq, France,
in 1986 and the Ph.D. degree in electrical engineer-
ing from the University of Lille, Villeneuve dAscq,
in 1990.
He is currently a Professor with the Ecole Centrale
de Lille, where he is also the Head of the Electri-
cal Engineering Department. He is the Leader of
the Power Electronic Research Group, Laboratoire
dElectrotechnique et dElectronique de Puissance
de Lille. His major elds of interest are hard switched power converters and
supercapacitors, particularly the control of multilevel topologies for medium-
and high-power applications with the aim of high power quality and high
efciency.