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

net/publication/270913915

Dynamical modeling procedure of a Li-ion battery pack suitable for real-time

applications

Article in Energy Conversion and Management · March 2015

DOI: 10.1016/j.enconman.2014.12.076

CITATIONS READS

61 1,135

4 authors, including:

Sandra Castaño-Solis Lucia Gauchia

Universidad Politécnica de Madrid Michigan Technological University
35 PUBLICATIONS 331 CITATIONS 43 PUBLICATIONS 658 CITATIONS

SEE PROFILE SEE PROFILE

Javier Sanz
University Carlos III de Madrid
68 PUBLICATIONS 983 CITATIONS

SEE PROFILE

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

Research visit to Université de Lille 1 (L2EP Lab) View project

Evaluation of Residential Battery Energy Storage Degradation in Smart Home using Detailed Hierarchical Bayesian Network View project

All content following this page was uploaded by Sandra Castaño-Solis on 18 December 2017.

The user has requested enhancement of the downloaded file.

 Dynamical modeling procedure of a Li-ion battery pack suitable for real-time
applications

S. Castano*1, L. Gauchia2, E. Voncila3, J. Sanz1

scastano@ing.uc3m.es, gauchia@mtu.edu, elena.voncila@ugal.ro, jsanz@ing.uc3m.es
1
Electric Engineering Department
Carlos III University of Madrid
28911 Leganes (Madrid) Spain
2
Department of Electrical and Computer Engineering
Michigan Technological University
49931 Houghton (Michigan) United States
3
Faculty of Automatic Control, Computer Science, Electrical and Electronics Engineering
Dunarea de Jos Universidad de Galati
Romania

*Corresponding author. Tel.: +34 916248851; Fax: +34 916249430

E-mail address: scastano@ing.uc3m.es (S. Castano)

Abstract- This paper presents the modeling of a 50 Ah battery pack composed of 56 cells,
taking into account real battery performance conditions imposed by the BMS control. The
modeling procedure starts with a detailed analysis of experimental charge and discharge
SOC tests. Results from these tests are used to obtain the battery model parameters at a
realistic performance range (20% - 80% SOC). The model topology aims to better describe
the finite charge contained in a battery pack. The model has been validated at three
different SOC values in order to verify the model response at real battery pack operation
conditions. The validation tests show that the battery pack model is able to simulate the real
battery response with excellent accuracy in the range tested. The proposed modeling
procedure is fully applicable to any Li-ion battery pack, regardless of the number of series
or parallel cells or its rated capacity.
Keywords- Li-ion battery, battery pack, dynamic, modeling, real-time.

I. INTRODUCTION
The evolution of the transportation and grid sectors towards a more sustainable future is
leading the research and development of electrochemical energy storage systems, and
batteries in particular. The relevance of batteries is increasing as new materials improve its
power and energy density, allowing batteries to take a more relevant role in vehicle
propulsion and smart grids. Lithium battery technology is currently dominant as it presents
high energy density, power density, and long shelf life [1]. However, it can present unstable
behavior that can potentially lead to thermal runway [23] [24]. This is the main reason why
lithium batteries need a battery management system (BMS). The BMS will protect the
battery from undervoltage/overvoltage, short-circuit (maximum current limit) and thermal
runway [2]. It will also measure cell voltage, module current, temperature, and estimate the
battery state-of-charge (SOC) and state-of-health (SOH). It is usual to set the SOC
operating limits to 20%- 80% in order to use battery efficiently [2], [13], and may be even
more restricted in some demanding applications [14].
Therefore, packs based on lithium technologies need a different modeling approach than
other technologies due to the presence and operation of the BMS. Battery models for
previous technologies still in use, such as nickel or lead, are exclusively based on the cell,
as no BMS is present [25]. This practice has been extended to lithium technologies [26],
[27] too, disregarding the effect of BMS on the battery operation.
This paper presents a different modeling approach that considers the effect of the BMS on
the battery operation. To frame this work with existing modeling techniques, a discussion
of the mainstream lithium batteries modeling approaches is presented next.
The most detailed models include electrochemical or physical based models [3]-[5], of a
single cell, which are able to accurately describe the internal phenomena taking place in the
battery. Despite their accuracy, these models are very complex, the coupled non-linear
differential equations that compose them are difficult to implement and require heavy
computational work [6], and they model a single cell instead of a whole pack and thus are
rarely used in real-time applications in BMS.
A common approach due to its simple design and fast computation is the electrical
equivalent circuit model, which can be developed for both cell and pack models. The
procedure to obtain these models can be through frequency or time domain tests. Frequency
domain tests, even if they can render highly detailed models, are highly time consuming,
need expensive equipment and are normally run on individual cells rather than on the whole
pack [6],[20]. These tests superpose a variable frequency signal to the battery current using
an impedance analyzer, which is an expensive equipment. This allows obtaining the battery
impedance over a wide frequency range, which gives insight to variable battery operation
depending on the dynamic requirements [33]. However, this is at the cost of long testing
times in order to reach the mHz frequency order. On the other hand, time domain tests can
result in less detailed models, but can be less time and resource consuming and can be used
to develop battery pack models. In this paper we will discuss models using time domain
tests.
Some authors have designed simple electrical circuit models composed by an ideal voltage
source in series with a constant internal resistance [6]-[8]. Such a model can only be used in
an early stage of battery sizing, because it does not offer information regarding the dynamic
behavior of battery [7], [9] and [21]. In order to take into account other variables, like the
state-of-charge (SOC), and electrochemical processes inside the battery, some improved
models have been proposed. Circuit models suggested by [9]-[11] introduce a RC network
to simulate battery kinetics and polarization process. More complex models have been
developed by adding electric passive and nonlinear components able to reproduce battery
runtime, non-linear processes and transient response [7], [12], [30] and [31].
However, all these models are focused on the short term dynamics on the battery and do not
take into account the operational limitations due to the BMS operation, as they consider
that the battery behavior is the same from 0% to 100 % SOC and do not take into account
the battery control restrictions present in practical applications. It is usual to set the SOC
operating limits between 20% to 80% in order to use battery efficiently [2], [13], and may
be even more restricted in some demanding applications [14]. Also, most authors only
experimentally validate the model for a single SOC. But the battery model may not be able
to correctly reproduce other SOC, especially those near the minimum and maximum SOCs.
The approaches used by these authors are only realistic if no BMS is involved. Lithium
technologies do include the BMS, which influences the battery operation. Hence, the
modeling approach must evolve into considering it.
A new approach that considers the BMS operation for this battery and its influence on the
pack performance is proposed here. This paper will demonstrate that the whole pack model
developed is able to represent the battery short and long term dynamics while considering
the BMS. The short term dynamics will include transient diffusion and charge transfer
process, whilst the long term will represent the long term dependence of the internal
voltage with respect to the SOC. This long term dependency of voltage with SOC will be
represented through a variable capacitor in order to better model the fact that a battery pack
has a limited capacity. This voltage-SOC dependency is commonly modeled through an
ideal voltage source [17]-[19], which fails to convey the finite nature of battery capacity.
The novel model proposed is experimentally validated in real-time at different values of
SOC, to prove that the pack model taking into account the BMS is accurate for short and
long term dynamics in the battery SOC operating range.
The contents of this paper are organized as follows. Section II presents the time domain
experimental testing procedure carried out to obtain the battery pack model. The
experimental data processing and analysis presented in Section III. These results are used to
carry out the model parameter fitting in Section IV. The model is experimentally validated
in real-time in Section V. Finally, Section VI presents the conclusions for this work.

II. EXPERIMENTAL TESTING PROCEDURE

The Li-ion battery pack used in our experiments consists of a commercial four parallel-
connected strings, each one composed by a series of seven groups of two cells in parallel
(in total, 56 MP-176065 Integration Saft cells), as shown in Fig. 1. There is a BMS control
board for each of the four parallel strings. The main battery characteristics are shown in
Table 1.
 Fig. 1. Battery layout

Rated voltage 25.9 V

Maximum voltage 29.4 V
Minimun cut-off voltage 20.3 V (aprox. 2.9 V /cell)
Capacity 50 Ah
Maximun current 50 A
Range of temperature (charge) -20ºC – 60ºC
Range of temperature (discharge) -30ºC – 55ºC
Table 1. Battery characteristics

The time domain tests start with the battery fully charged. It is then discharged under
pulsed current, followed then by a charging process. The details for each test are the
following:
Full charge: The battery is initially charged to 100% by using a constant current/constant
voltage (CC/CV) charging method, with I = 25 A, and final voltage U = 29.4 V.
Discharge test: The battery pack is discharged at regular intervals with current pulses of 10
A during 30 minutes followed by 90 minutes of relaxation time. The battery pack open
circuit voltage (OCV) is measured at the end of each relaxation period.
Charge test: After completing the discharge test, the battery is charged again at regular
intervals with current pulses of 10 A during 30 minutes followed also by 90 minutes of
relaxation time. The cut-off voltage during the charging process is set at 29.4 V. The OCV
is also measured at the end of each relaxation period.
The experimental setup used for these tests is depicted in Fig. 2, where the experiment
control and the real-time data acquisition were implemented on a real-time dSpace system.
A Chroma 80V/300A electronic load and a Sorensen SGI 60V/167A power source were
used to test the battery pack. The voltage and current during the tests were measured by
isolated voltage and current sensors to provide an adequate galvanic isolation.
 Fig. 2. Experimental setup

The general results for these tests are depicted in Fig. 3. As it can be observed, 10A
discharge and charge pulses were applied. However, the charging current pulses are limited
when the battery voltage reaches 29.4V. This effect is due to the BMS, which protects the
battery pack against overcharge. This battery operation constraint imposed by the BMS is
necessary as lithium-ion batteries are prone to thermal runway. The final voltage curves
presented in Fig. 3 differ from the rest due to the current limitation imposed by the BMS.
To study this effect more closely, each voltage curve is extracted from Fig. 3 and plotted
independently. All the curves are plotted together in Fig. 4 to allow a better comparison.

Fig. 3. Discharge (left) and charge (right) test results

As shown in Fig. 4, the initial non-linear voltage evolution for each curve varies with the
SOC. The voltage slope decreases for decreasing SOC during discharge, and increases for
increasing SOC during charge. There is an exception for this last statement, as for SOC
above 87% during charge the voltage slope decreases again. This contradictory effect is due
to the limitations imposed by the BMS.
Based on these results, two conclusions can be made. The first one is that the real battery
operating range is between 20-80% SOC, as beyond this range the battery will be unable to
meet the current requirements needed by the application. The second one is that the voltage
slopes vary with the battery SOC. It is relevant to reflect in the battery model this slope
variation seen in the different voltage curves in Fig. 4, as taking into account this fact is key
for a good modeling accuracy. Failing to do so will result in the battery model drifting
when working at different SOCs. It is then pertinent to validate the model during the whole
SOC interval, and to the best knowledge of the authors, this has not been thoroughly done
by other work present in the scientific literature. This effect will be considered during the
data processing and analysis presented in the next section.

Fig. 4. Voltage evolution for each step: discharge (left) and charge (right)

III. DATA PROCESSING AND ANALYSIS

In this Section, the experimental results obtained in Section II are processed in order to
work towards the battery pack model. Therefore, each of the voltage curves presented in
Fig. 4, both for charge and discharge, will be processed and analyzed following the
procedure explained below. For the sake of readability, this Section will present the data
processing for one curve. The discharge and charge curves taken as an example are
presented in Fig. 5.
Ubat(V)

Ubat(V)

Fig. 5. Zoom of the discharge (left) and charge (left) tests

It can be seen that both discharge and charge processes share a common evolution, in which
four different phenomena can be identified, and which are labeled zone 1 to 4 in Fig. 5.

Zone 1: An abrupt voltage variation due to the battery series resistance, which causes a
voltage vertical drop during discharge and a vertical step during charge.

Zone 2: A non-linear voltage decrease (for discharging) or increase (for charging)

due to the variations of electric charge during the current flow. This variation will
only take place when current flows.
Zone 3: Again, an abrupt voltage variation due to the battery series resistance.
Zone 4: A non-linear transient voltage recovery due to the diffusion processes
associated to the relaxation periods after the current is suddenly cut-off.
It can be observed in Fig. 5, and as explained in the previous Section, zone 2 can present
different decreasing (discharge) or increasing (charge) rates, as seen in Fig. 4. This
variation can be considered as the superposition of two components: a linear variation
basically related to the charge depletion/accumulation processes, and a non-linear one that
is related with the charge transfer and diffusion processes inside the battery. This last can
be easily fitted to an exponential variation, as it is shown in Fig. 6, where the discharge case
is analyzed. An equivalent analysis has been carried out for the charge process, but is not
presented here.

Fig. 6. Decomposition of zone 2 during discharge, shown in Fig. 5

Each one of the zones 1-4 seen in Fig. 5 can be represented and explained by means of the
proposed equivalent circuit depicted in Fig. 7. The abrupt voltage drops are due to the
series resistance (R0), the exponential curve from Fig. 6 is associated with the Rt-Ct parallel
network. The internal voltage can be represented by a constant ideal voltage source (E0) in
series with a variable capacitor of very large capacitance (C), representing the voltage
variation due stored/depleted charge. This battery model splits the effect of the variation of
the stored charge on the internal voltage of the battery into two components: E0 and C. To
represent this phenomenon, some authors [12] have proposed the use of voltage-controlled
or current-controlled sources. Others works use a constant capacitor to model the cell SOC
[30] or the variation in open circuit voltage [31]. However, a variable capacitor (though of
an amazingly high capacitance, indeed) better reflects the fact that the battery pack has a
limited capacity, and that its charge is time- and current dependent. This capacitor will
serve to represent the linear component of the voltage drop described before and depicted in
Fig. 6.

Fig. 7. Proposed battery pack equivalent circuit

IV. PARAMETER FITTING

This Section will present the methodology to obtain the equivalent circuit parameters and
its dependency on the battery SOC and charge/discharge process. This has been done for
the charge and discharge cycles in Fig. 4 in the 20-80% interval. Fig. 8 depicts, for a single
discharge profile, the voltage values used to calculate the parameters for a single discharge
curve. This is repeated for the rest of discharge and charge curves.
 Fig. 8. Battery voltage values for a typical discharge profile

A. Measurement of E0
The model voltage source represents the initial battery open circuit voltage, which is due to
the standard potentials of each electrode. This voltage is only measureable at the beginning
of the battery cycle, and is then used to represent E0. This value is kept constant throughout
the battery simulation, as it cannot be updated online. In practical implementations of the
model, the open circuit voltage measured can only be updated after the battery pack has
remained in a long relaxation period.

B. Calculating C
This work proposes modeling the voltage variations due to changes in the battery charge by
means of an ideal capacitor. The use of an ideal capacitance in series with a constant ideal
voltage source allows adding to the model the limitations of a real battery capacity, rather
than using a constant or controlled voltage source as other previous models do [12]. This
capacitance C is obtained from the parameter α, where α represents the slope between the
voltage (ΔU) and charge variation (ΔQ), as seen in (1), and is inversely proportional to the
capacitance C.

1 ΔU u1d − u6 d
α= = = (1)
C ΔQ i ⋅ Δt

The parameter α is calculated in (2) considering the difference of the open circuit voltages
between the beginning and the end of the each voltage curve. These values can be extracted
from Fig. 8. The term Δt is the time interval during which the current is applied (0.5 h).

Fig. 9 presents the evolution of α during charge and discharge for states-of-charge (SOC)
ranging from 20 to 80%. As explained above, the reason for choosing this SOC range is to
account for the usual operational limits the BMS imposes to the battery. As it can be seen,
the values for α follow a similar evolution during charge and discharge processes, albeit
with higher values during charge. This phenomenon can be related to the battery hysteresis
effect, as α is calculated using the open circuit voltage (OCV) measured after 90 min. The
OCV is highly dependent on the SOC and the charge/discharge process due to the different
charge concentration at the electrode surface [16]. The model can then either include two
different α values for charge and discharge, or just their average value. In this case, the
linear regression curve between average values of α and SOC, obtained from data in Fig. 9
using the lineal regression function of the software Statghraphics® (version Centurion
XVI), is given in (2).

α = −3.4128 ⋅10−6 + 4.3995 ⋅10−7 ⋅ SOC (2)

Fig. 9. α vs. SOC for discharge, charge and average

C. Calculating R0
The ohmic voltage drop in a battery is due to its internal resistance, which is a lumped
parameter which reflects the various contact resistances of the current collectors, as well as
the ions movement in the electrolyte, among others. The resistance R0 is calculated both for
charge (R0c) and discharge (R0d). It is also calculated twice during each cycle, as the internal
resistance effect is visible when the current is switched on (R0d1) or off (R0d2). These effects
are related to zones 1 and 3 in Fig. 5. These voltage drops are computed separately for
charge and discharge cycles, and averaged afterwards. They are calculated as shown in (3)
and (4), by taking the experimental values from Fig. 8.
u −u
R0 d 1 = 1d 2 d (3)
i
u5d − u4 d
R0 d 2 = (4)
i

Fig. 10 presents the evolution of series resistance for different SOCs. It can be observed
that the resistances calculated during a voltage drop for discharge, increases from 0.049
ohm to 0.053 ohm for values under 60% SOC. Whilst during the voltage boosts calculated
during the charging process it follows a decreasing evolution until 70 % SOC. The
deviations between the charge and discharge values are below 10%, so it is possible to
calculate an average resistance that varies around 0.050 ohm.

0.054
R0 discharge
0.053
R0 charge
0.052 R0 average

0.051
R0 (ohm)

0.05

0.049

0.048

0.047

0.046

0.045
20 30 40 50 60 70 80
SOC (%)

Fig. 10. R0 vs. SOC for discharge, charge and average

D. Calculating Rt-Ct
The RC network to be calculated represents the exponential behavior produced by
electrochemical processes (charge transfer and diffusion) inside the battery. This voltage
variation is described as (5).

1
U RC = ∫ ⋅ ( i − iRt ) ⋅ dt
Ct
(5)
Where Rt and Ct are calculated by identifying the correct voltage values, which is done for
all of the curves presented in Fig. 4. An example for the discharge case (Rtd, Ctd) is given in
(6) and (7), where the voltages are identified in Fig. 8. Where τ represents the time
constant.
u − u3d
Rtd = 2 d
i (6)
τ
Ctd =
Rtd (7)
The evolution of the Rt-Ct network parameters with the SOC for charge and discharge is given in
Fig. 11. It can be observed that the parameters have similar evolution for both charge and
discharge, and can be modeled separately or using an average value. The Rt and Ct values
are heavily dependent on the nonlinear voltage evolution. As presented earlier in Fig. 4,
each curve represents the voltage evolution for different SOC values. Each voltage curve
presents a different slope for the non-linear region (called zone 2 in Fig. 5). The variations
in the slope will directly affect the Rt value, and therefore the Ct value.
Rt (ohm)

Ct (F)
Fig. 11. Rt (left) and Ct (right) evolution with SOC for discharge, charge and average

Once again, the average values Rt and Ct between charge and discharge represented in Fig.
11 can be expressed as a function of SOC, using the polynomial regression function of
Statgraphics® (version Centurion XVI), yielding equations (8) and (9).

Rt = 0.0436 − 0.0016 ⋅ SOC + 3.6071⋅10−5 ⋅ SOC 2 − 2.5098 ⋅10−7 ⋅ SOC 3 (8)

Ct = 11741.6 − 379.131 ⋅ SOC + 5.3713 ⋅ SOC 2 − 0.0221 ⋅ SOC 3 (9)

The battery model is shown in Fig. 12, where the battery terminal voltage Ubat is defined by
equation (10). The current i, the state of charge SOC and E0 corresponds to the model
inputs, and the battery voltage Ubat represents model output. The capacitance C is
calculated as 1/α. The model is able to reproduce the battery terminal voltage between the
operating range of 20-80% SOC, which is imposed by the battery management system.

Fig. 12. Battery pack model

1 1 ⎛ ΔU Ct ⎞
U bat = Eo −
C ∫ i ⋅ dt − Ro ⋅ i − ∫ ⋅ ⎜ i −
Ct ⎝
⎟ ⋅ dt
Rt ⎠
(10)
 V. REAL-TIME MODEL VALIDATION
In order to prove that the model is able to reproduce the voltage battery response in a real-
time application, a smaller scale electric vehicle power is simulated over a New European
Driving Cycle (NEDC) using a dSpace® real-time data acquisition system (version 6.2)
and the dSpace® toolboxes of Matlab/Simulink® (version 2007b). The driving cycle is
depicted in Fig. 13.
Speed (kph)

Fig. 13. New European Driving Cycle (NEDC)

The electric vehicle is simulated using hardware-in-the-loop simulation (HIL), in which a

real device emulates the vehicle power demand [28], [29]. In this experimental test bench,
the devices used to emulate the vehicle are an electronic load and a power source, which
are controlled in synchronization to simulate the vehicle power demand during acceleration
and braking. The real battery pack is tested under the power requirements of the electric
vehicle simulated as HIL following a NEDC. The experimental setup used is presented in
Fig. 14, where the upper part refers to the power circuit and the lower part to the HIL
control.
By means of the dSpace® real-time system, both the vehicle HIL and the battery are
controlled. For this purpose, the vehicle power demand is calculated according to the
vehicle NEDC speed profile. Knowing that the real battery pack imposes its voltage to the
simulated HIL vehicle, the electronic load and power source are externally programmed to
supply/absorb the current demanded through an external voltage control signal. This
voltage control signal is processed by the real-time OS in order to adapt it to the levels
required by each device.
As a result of this real-time HIL simulation, a current profile and an experimental battery
voltage are obtained. The simulations are carried out at 25%, 50% and 75% SOC to prove
that the model is accurate in the battery SOC operating range.
 A
ELECTRONIC LOAD V BATTERY

Measurement
signals
POWER SOURCE

PC
dSPACE hardware
Control signals and Real‐time OS Optical cable

Vehicle
resistive
Vehicle forces
dSpace driving + Power
Vehicle Transmission demand
real‐time cycle
simulation system
OS
+
Electric drive

Voltage adaptation

dSpace
I/O
hardware

Battery voltage and Power source Electronic load

current from sensors control signal control signal

Fig. 14. Experimental setup for real-time simulation

The vehicle experimental current demand obtained earlier (Fig. 15) is applied to the battery
model developed in Matlab/Simulink®. The objective is to experimentally validate the
battery model in the whole SOC operating range. For this reason, the model is tested at the
same SOCs values (25%, 50% and 75%) than the experimental setup. Figs. 16-18 show the
comparison of the simulated and experimental voltage of the battery. As depicted in Fig.
19, the maximum voltage deviation is less than 400 mV and the maximum error is less than
1.5% for each SOC tested. Therefore, the proposed battery pack model is highly accurate
for the three analyzed cases. Consequently, this model is consistently accurate enough to
simulate throughout the whole battery operational SOC interval.
 Fig. 15. Battery current demand over a NEDC
Voltage (V)

Fig. 16. Battery experimental vs. simulated voltage at 25% SOC

 28.5
Ubatexp
28 Ubatmod

27.5

27

26.5

26

25.5

25

24.5

24
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Time (h)

Fig. 17. Battery experimental vs. simulated voltage at 50% SOC

Voltage (V)

Fig. 18. Battery experimental vs. simulated voltage at 75% SOC

 1
25% SOC
0.8

Deviation (V)
50% SOC
0.6 75% SOC
0.4
0.2
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Time(h)
2

1.5
Error (%)

0.5

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Time (h)

Fig. 19. Error between the battery model and experimental values

In order to compare the results of this model with other circuits, an additional comparison is
made. As explained before, circuits used in real-time applications must have a low
computational load. For this reason, the so-called PNGV model [30], [32] is chosen as a
comparison. In this circuit, a constant capacitor is used to model the cell SOC. In order to
adapt this model to the tested battery pack the capacitor C2 is calculated by equation (11).
Kpack represents the rated capacity of the pack and V100%SOC and V0%SOC are the voltage at
100% SOC and 0%SOC respectively. The rest of the circuit parameters remain the same.

3600( s ) ⋅ K pack ( Ah)

C2 = (11)
V100% SOC − V0% SOC

Figs. 20-22 show the comparison of the simulated voltages and experimental voltage of the
battery pack. Here, Ubmod 1 is the simulated voltage of the model proposed in this paper
and Ubmod 2 is the simulated voltage of the well-known PNGV circuit used for
comparison. As it can be seen the proposed model reproduces the experimental voltage
response of the battery pack better than the PNGV model. The maximum error of the
PNGV model is around 3%, whilst the proposed model presents a maximum error less than
1.5%. Only in the case of the operation at 75% SOC both models present a similar
accuracy. This fact demonstrates the importance of validating the model for the whole SOC
range.
 28
Ubexp
27.5 Ubmod 1
Ubmod 2

27

26.5
Voltage (V)

26

25.5

25

24.5

24

23.5
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Time (h)

Fig. 20. Battery experimental vs. simulated voltages at 25% SOC

28.5
Ubexp
Ubmod 1
28
Ubmod 2

27.5

27
Voltage (V)

26.5

26

25.5

25

24.5

24
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Time (h)

Fig. 21. Battery experimental vs. simulated voltages at 50% SOC

 29.5
Uexp
Umod 1
29 Umod 2

28.5

28
Voltage (V)

27.5

27

26.5

26

25.5
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Time (h)

Fig. 22. Battery experimental vs. simulated voltages at 75% SOC

VI. CONCLUSIONS
The aim of this paper is to present the experimentally validated model of a real Li-ion
battery pack in the whole SOC operating range established by the BMS control. It also
improves the model topology to consider a more realistic relation between the finite battery
capacity and voltage.
The battery model parameters evolution with SOC was investigated. The parameters were
specifically defined to represent the battery pack real performance whilst limiting the model
complexity. This allows the model to be able to run in real-time, matching the experimental
results. The model is able to represent the battery ohmic voltage drop that reproduces the
effect of the battery internal resistance, the transient voltage due to diffusion and transfer
charge processes within the battery pack, and a third component that accounts for the
variation of the internal battery voltage due to SOC changes. The battery model topology
was specifically modified to better represent the battery internal voltage, which can
significantly affect the model accuracy. A capacitor was added to include this effect, by
means of a parameter (α) that is the inverse of the capacitance and represents the
relationship between charge (Q) and voltage. This approach represents in a realistic way the
limited battery capacity, which depends on SOC and current rate.
The battery model was experimentally validated for an electric vehicle real-time
application. The model was validated, not only for a single SOC, as it is normally found in
literature, but for the whole SOC operating range: from 20 to 80% SOC. Also, an additional
comparison with the so-called PNGV model was realised. The model was consistently
accurate and robust across the whole operating range.
 VII. REFERENCES

[1] A. Khaligh, L. Zhihao, "Battery, Ultracapacitor, Fuel Cell, and Hybrid Energy Storage
Systems for Electric, Hybrid Electric, Fuel Cell, and Plug-In Hybrid Electric Vehicles:
State of the Art," IEEE Transactions on Vehicular Technology, vol.59, no.6, pp. 2806-
2814, 2010.

[2] S. Lukic, J. Cao, R. Bansal, F. Rodriguez, A. Emadi, “Energy Storage Systems for
Automotive Applications,” IEEE Transactions on Industrial Electronics, vol. 55, No.
6, pp.2258-2267, 2008.

[3] Y. Hu, S. Yurkovich, Y. Guezennec, and B.J. Yurkovich, “A technique for dynamic
battery model identification in automotive applications using linear parameter varying
structures,” Control Engineering Practice, vol. 17, no. 10, pp. 1190-1201, 2009.

[4] K. Smith, C. Rahn, C. Wang, "Model-Based Electrochemical Estimation and

Constraint Management for Pulse Operation of Lithium Ion Batteries," IEEE
Transactions on Control Systems Technology, vol.18, no.3, pp.654-663, 2010.

[5] G. Sikha, R. White, and B. Popov, “A mathematical model for a Lithium-Ion

Battery/Electrochemical Capacitor Hybrid System,” Journal of the Electrochemical
Society, vol. 152, no. 8, pp. 1682 – 1693, 2005.

[6] A. Shafiei, A. Momeni, and S. Williamson, “Battery Modeling Approaches and

Management Techniques for Plug-in Hybrid Electric Vehicles,” Vehicle Power and
Propulsion Conference, VPPC’11, pp 1-5, 2011.

[7] S. Abu-Sharkh, D. Doerffel, “Rapid test and non-linear model characterisation of solid-
state lithium-ion batteries”, Journal of Power Sources, vol. 130, pp. 266–274, 2004.

[8] A. Purvins, I. T. Papaioannou, L. Debarberis, “Application of battery-based storage

systems in household-demand smoothening in electricity-distribution grids,” Energy
Conversion and Management, vol.65, pp 272-284, 2013.

[9] H. He, R. Xiong, H. Guo, S. Li, “Comparison study on the battery models used for the
energy management of batteries in electric vehicles,” Energy Conversion and
Management, vol. 64, pp.113-121, 2012.

[10] L. Gao, S. Liu, and R.A. Dougal, “Dynamic Lithium-Ion Battery Model for System
Simulation,” IEEE Transactions on components and packaging technologies, vol. 25,
no. 3, pp 495-505, 2002.

[11] M. Dubarry, B. Y. Liaw, “Development of a universal modeling tool for rechargeable

lithium batteries,” Journal of Power Sources, vol. 174, pp 856-860, 2007.
[12] M. Chen, G.A. Rincon-Mora, “Accurate Electrical Battery Model Capable of
Predicting Runtime and I–V Performance”, IEEE Transaction on energy conversion,
vol. 21, no. 2, 2006.

[13] N.C. Nair, N. Garimella, “Battery energy storage systems: Assessment for small-scale
renewable energy integration,” Energy and Buildings, vol. 42, pp 2124-2130, 2010.

[14] R.F. Nelson, “Power requirements for batteries in hybrid electric vehicles,” Journal of
Power Sources, vol. 91, pp 2-26, 2000.

[15] S. Kim, “The novel state of charge estimation method for lithium battery using sliding
mode observer.” Journal of Power Sources, vol. 163, pp 584-590, 2006.

[16] S. Piller, M. Perrin, A. Jossen, “Methods of state-of-charge determination and their

applications.” Journal of Power Sources, vol. 96, pp 113-120, 2001.

[17] G. Plett, “Extended Kalman filtering for battery management systems of LiPB-bases
HEV battery packs. Part 2. Modeling and identification.” Journal of Power Sources,
vol. 134, pp 262-276, 2010.

[18] S. Lee, J. Kim, J. Lee, B.H. Cho, “State-of-charge and capacity estimation of lithium-
ion battery using a new open-circuit voltage versus state-of-charge.” Journal of Power
Sources, vol. 185, pp 1367-1373, 2008.

[19] H. Dai, X. Wei, Z. Sun, J. Wang, W. Gu, “Online cell SOC estimation of Li-ion battery
packs using a dual time-scale Kalman filtering for EV applications.” Applied Energy,
vol. 1, no. 95, pp 227-237, 2012.

[20] A. Eddahech, O. Briat, N. Bertrand, J. Deletage, J. Vinassa, “Behavior and state-of-

health monitoring of Li-ion batteries using impedance spectroscopy and recurrent
neural networks,” International Journal of Electrical Power and Energy Systems,
vol.42, pp 487-494, 2012.

[21] H.L. Chan, D. Sutanto, “A New Battery Model for use with Battery Energy Storage
Systems and Electric Vehicles Power Systems,” Power Engineering Society Winter
Meeting, vol. 1, pp. 470 – 475, 2000.

[22] J. Vetter, P. Novák, M.R. Wagner, C. Veit, K.C. Möller, J.O. Besenhard, M. Winter,
M. Wohlfahrt-Mehrens, C. Vogler, A. Hammouche, “Ageing mechanisms in lithium-
ion batteries,” Journal of Power Sources, vol. 147, pp 269-281, 2005.

[23] D. Andrea, “Battery Management Systems for large Lithium-Ion Battery Packs,”
Artech House, Boston, 2010.

[24] H. Dai, X. Zhang, , X. Wei, , Z. Sun, , J. Wang, F. Hu, “Cell-BMS validation with a
hardware-in-the-loop simulation of lithium-ion battery cells for electric vehicles,”
International Journal of Electrical Power and Energy Systems, vol. 52, pp 174-184,
2013.
 [25] L. Gauchia, J. Sanz, “Per unit representation of electrical magnitudes in batteries: A
tool for comparison and design”. Energy Conversion and Management, vol. 50, pp
554-560, 2009.

[26] E. Vinot, R. Trigui, “Optimal energy management of HEVs with hybrid storage
system”. Energy Conversion and Management, vol. 76, pp 437-452, 2013.

[27] C. T. Chung, Y. H. Hung, “Energy improvement and performance evaluation of a

novel full hybrid electric motorcycle with power split e-CVT,” Energy Conversion and
Management vol. 86, pp 216-225, 2014.

[28] A. Bouscayrol, “Different types of hardware-in-the-loop simulation for electric drives",

IEEE International Symposium on Industrial Electronics, June 2008, Cambridge, UK.

[29] L. Gauchia, J. Sanz, “A per-unit hardware-in-the-loop simulation of a fuel cell/battery

hybrid energy system”, IEEE Transactions on Industrial Electronics, vol. 57, no.4, pp.
1186-1194, 2010.

[30] C. Antaloe, J. Marco, F. Assadian, “A novel method for the Parametrization of a Li-Ion
Cell Model for EV/HEV Control Applications”, IEEE Transactions on Vehicular
Technology, vol. 61, no.9, pp. 3881-3891, 2012.

[31] M. Daowd, N. Omar, B. Verbrugge, P. Van Den Bossche, J. Van Mierlo, “Battery
Models Parameter Estimation based on MATLAB/Simulink®. EVS-25 Shenzhen,
China, 2010.

[32] M. Al Sakka, H. Gualous, N. Omar, J. Van Mierlo, “Batteries and Supercapacitors for
Electric Vehicles, New Generation of Electric Vehicles”, InTech, 2012, DOI:
10.5772/53490.

[33] D. Andre, M. Meiler, K. Steiner, H. T. Walz, Soczka-Guth, D.U. SAUER,

Characterization of high-power lithium-ion batteries by electrochemical impedance
spectroscopy. I. Experimental investigation. Journal of Power Sources, vol. 196, no.
12, pp. 5334-5341, 2011

View publication stats