Preprints,

Advances 8th IFAC International

in Automotive Control Symposium on
Preprints, 8th IFAC International Symposium on
Advances
Preprints,
Preprints,
June 19-23,in2016.
8th
8thAutomotive
IFAC Control
International
IFACNorrköping,
International Symposium
Symposium
Sweden on
ononline at www.sciencedirect.com
Advances in Automotive Control Available
June 19-23,in
Advances
Advances in2016. Norrköping,
Automotive
Automotive Sweden
Control
Control
June 19-23, 2016. Norrköping, Sweden
June
June 19-23,
19-23, 2016.
2016. Norrköping,
Norrköping, Sweden
Sweden
ScienceDirect
A
A Study
Study on
on Battery Model Parametrisation
IFAC-PapersOnLine 49-11 (2016) 048–053
Battery Model Parametrisation Problem
Problem ––
A
A Study on Battery Model Parametrisation Problem ––
A Study
Study on
on Battery
Application-Oriented
Battery
Application-Oriented Model
Model Parametrisation
Trade-offs
Trade-offs between Accuracy
Parametrisation
between Accuracy Problem
and
Problem
and Simplicity
–
Simplicity
Application-Oriented
Application-Oriented Trade-offs
Trade-offs between
between Accuracy
Accuracy and
and Simplicity
Simplicity
Application-Oriented
Abbas Fotouhi*,Trade-offs
Daniel J. Auger*, between
Karsten Accuracy and
Propp*, Stefano Simplicity
Longo*
Abbas
Abbas Fotouhi*,
Fotouhi*, Daniel
Daniel J.
J. Auger*,
Auger*, 
Karsten
Karsten Propp*,
Propp*, Stefano
Stefano Longo*
Longo*
Abbas Fotouhi*, Daniel J. Auger*,  Karsten Propp*, Stefano
Stefano Longo*
* School ofAbbas Aerospace,Fotouhi*, Danieland
Transport J. Auger*,
Manufacturing,
 Karsten Propp*,University,
Cranfield Longo*
MK43 0AL, UK.
*
* School
School of
of Aerospace,
Aerospace, Transport
Transport and
and Manufacturing,


Manufacturing, Cranfield
Cranfield University,
University, MK43
MK43 0AL,
0AL, UK.
UK.
(e-mails: a.fotouhi@cranfield.ac.uk;
* School of Aerospace, d.j.auger@cranfield.ac.uk;
Transport and Manufacturing, k.propp@cranfield.ac.uk;
Cranfield University, MK43s.longo@cranfield.ac.uk
0AL, UK. )
(e-mails:
(e-mails: a.fotouhi@cranfield.ac.uk;
* School of Aerospace,
a.fotouhi@cranfield.ac.uk; d.j.auger@cranfield.ac.uk;
Transport and Manufacturing,
d.j.auger@cranfield.ac.uk; k.propp@cranfield.ac.uk;
Cranfield University,
k.propp@cranfield.ac.uk; MK43s.longo@cranfield.ac.uk
0AL, UK.
s.longo@cranfield.ac.uk )))
(e-mails: a.fotouhi@cranfield.ac.uk; d.j.auger@cranfield.ac.uk; k.propp@cranfield.ac.uk;
(e-mails: a.fotouhi@cranfield.ac.uk; d.j.auger@cranfield.ac.uk; k.propp@cranfield.ac.uk; s.longo@cranfield.ac.uk ) s.longo@cranfield.ac.uk
Abstract: This study is focused on fast low-fidelity battery modelling for online applications. Because
Abstract: This
Abstract: This study
study is is focused
focused on on fast
fast low-fidelity
low-fidelity battery battery modelling
modelling for for online
online applications.
applications. Because Because
the battery This
Abstract: parameters
study change
is focused dueon to fast
variations of battery’s
low-fidelity battery states,
modellingthe model for may need
online to be updated
applications. Because dur-
the
the battery
Abstract:
battery parameters
This study
parameters change
is focused
change due
due to
on
to variations
fast of
low-fidelity
variations of battery’s
battery
battery’s states,
modelling
states, the
the model may need
foridentification
model online
may need to be
applications.
to be updated
updatedBecause dur-
dur-
ing
the operation. This can be achieved through the use of an online parameter technique, mak-
ing battery
the operation.
battery parameters
This can
parameters canchange
change due
be achieved
achieved to variations
through
duemeasurements.
tothrough
variationstheof usebattery’s
of of an states,
an online
battery’s online the
parameter
states,parameter model
the algorithm’s may
may need
identification
modelidentification
need to technique,
be
bea updated dur-
mak-
ing
ing operation.
use of online
operation. This
This can be
current-voltage
be achieved through the
the use
The
use of
ofparametrisation
an online parameter speedtoistechnique,
identification
updated
crucial mak-
technique,
dur-
issue
mak-
ing
ing use
use of
operation.
of online
This
online current-voltage
can be
current-voltageachieved measurements.
through
measurements. the The
use
The ofparametrisation
an online
parametrisation parameter algorithm’s speed
identification
algorithm’s speed is a crucial
technique,
is a crucial issue
mak-
issue
in such
ing use of applications.
of This paper describes
online current-voltage
current-voltage measurements.a study The exploring the trade-offs
parametrisation between
algorithm’s speed speed and
is aaand accuracy,
crucial issue
in such
ing
in such
use applications.
online
applications. This paper
paper describes
describes
measurements.a study
studyThe exploring the trade-offs
parametrisation trade-offs between
algorithm’s speed speed is accuracy,
crucial issue
considering
in such equivalentThis
applications. circuit
This paper models withaadifferent
describes study exploring
levels
exploring ofthe
thecomplexity
trade-offs andbetween
different
between speed
speed and accuracy,
parameter-fitting
and accuracy,
considering
in such
considering equivalent
applications.
equivalent circuit
This paper
circuit models
models with
describes
with adifferent
study
different levels
exploring
levels of
of thecomplexity
trade-offs
complexity and different
between speed parameter-fitting
and accuracy,
algorithms. A
considering visual investigation
equivalent circuit models of the
withbattery
differentparametrisation
levels of problem and is alsodifferent
proposed parameter-fitting
by obtaining
algorithms. A
considering
algorithms.
battery model
A visual investigation
equivalent
visual investigation
identificationcircuitsurfaces
models of the
of the
which
battery
withbattery
different
help
parametrisation
levels
parametrisation
us to avoidof complexity
problem and
complexity
problem
unnecessary
is also
and
is
different
also proposed
different
proposed
complexities.
parameter-fitting
by obtaining
obtaining
parameter-fitting
Threeby standard
algorithms.
battery
algorithms. model A
A visual
visual investigation
identification surfaces
investigation of
of the
which
the battery
help
battery parametrisation
us to avoid
parametrisation problem
unnecessary
problem is
is also proposed
complexities.
also proposed Threeby
by obtaining
standard
obtaining
battery
fitting
battery model
algorithms
model identification
are used
identification tosurfaces
parametrise which help
battery us to
models avoid
using unnecessary
current-voltage complexities.
measurements. Three standard
For each
fitting algorithms
battery
fitting algorithms
model are used
used to
identification
are tosurfaces
parametrise
surfaces
parametrise
which
which help
help us
battery
battery us to
models avoid
usingunnecessary
to evaluated
models avoid
using current-voltage
unnecessary
current-voltage
complexities.
measurements.
complexities.
measurements.
Three
Three standard
For
standard
For each
each
level
fitting of complexity, the algorithms performances are using experimental data from a small
level of
fitting ofalgorithms
complexity, are used
the to
to parametrise
algorithms battery
performances models using
using current-voltage
are evaluated
evaluated using experimental
experimental measurements.
data from fromFor each
aaa small
small
NiMHofalgorithms
level
level complexity,
battery pack.
complexity,
areAn used
the
the
parametrise
algorithms
application-oriented
algorithms
battery
performances
performancesview models
onare
this
are trade-offs
evaluated
current-voltage
using
is discussed
using experimental
measurements.
whichdata demonstrates
data from
For each
that
small
NiMH
level
NiMH ofbattery
battery pack.
complexity,
pack. An
Anthe application-oriented
algorithms
application-orientedperformancesview
view on
on this
are trade-offs
evaluated
this trade-offs is
using
is discussed
experimental
discussed which
which demonstrates
data from
demonstrates a that
small
that
the
NiMH final target of the battery parametrisation problem can significantly affect the choice of the fitting al-
the
NiMH
the finalbattery
final target pack.
battery
target pack.
of theAn
of the An application-oriented
battery parametrisation view
application-oriented
battery parametrisation problem
view
problem
on
on this
cantrade-offs
this
can significantly
trade-offs
significantly
is
is discussed
affect the
discussed
affect thewhich
choicedemonstrates
which
choice of the
demonstrates
of fittingthat
the fitting al-
that
al-
gorithm
the and
final and
targetbattery
of the model
themodel
battery structure.
parametrisation problem problem can can significantly
significantly affect affect the the choice
choice of of the
the fitting
fitting al-al-
gorithm
the
gorithmfinal target
and battery
of
battery battery
model structure.
parametrisation
structure.
gorithm
©
gorithm
Keywords: and
2016, IFAC battery
andBattery
battery model
model structure.
(International
Modelling, Federation
structure. of Automatic
Parametrisation, Control) Hosting
Identification, Algorithm,by Elsevier
Electric Ltd.Vehicle.
All rights reserved.
Keywords: Battery
Keywords: Battery Modelling,
Modelling, Parametrisation,
Parametrisation, Identification,
Identification, Algorithm,
Algorithm, Electric Electric Vehicle.
Vehicle.
Keywords: Battery
Keywords: Battery Modelling,
Modelling, Parametrisation,
Parametrisation, Identification,
Identification,
 Algorithm, Electric
Algorithm, Electric Vehicle.
Vehicle.

 an estimator to predict battery states, Fotouhi et al. (2015).
1. INTRODUCTION  an estimator to predict battery states, Fotouhi et et al.
al. (2015).
(2015).
1. INTRODUCTION an
an estimator
System
estimator to
identification
to predict
predict battery
technique
battery is usedFotouhi
states,
states, in this framework
Fotouhi et al. as
1.
1. INTRODUCTION System
an estimatoridentification
to predict technique
battery for is used
states, in
Fotouhi this al. (2015).
framework
etparametrisa-
(2015). as
Hybrid electric vehicles 1. INTRODUCTION
(HEVs) are well-established in the System
INTRODUCTION one
System
one
of
of
the
the
identification
existing
identification
existing
technique
approaches
technique
approaches
is used
is
for
online
used
online
inmodel
in
this framework
this
model framework
parametrisa-
as
as
Hybrid electric vehicles (HEVs) are well-established in the System
one
tion of aidentification
the existing approaches
time-varying technique
system. ECN is used
for online inmodel
battery this framework
model parametrisa-
parametri- as
Hybrid
market
Hybrid electric
and
electric vehicles
electric vehicles
vehicles (HEVs)
(EVs)are
(HEVs) arewell-established
are growing in popularity.
well-established in
in the
the one
tion of the
a existing
time-varying approaches
system. for
ECN online
battery model model parametrisa-
parametri-
market
Hybrid and electric
electric vehicles
vehicles (EVs)
(HEVs) are
are growing in
well-established popularity.
in the one
tion
sationof
of the
a existing
time-varying
problem can approaches
be system.
classified for
ECN as online
anbattery model model
identification parametrisa-
parametri-
problem
market
This and
trend iselectric
likely vehicles
to sustain(EVs) for the are growing
foreseeable in popularity.
future. De- sation tion ofofproblem
market
This
market and
trend
andof iselectric
likely
electric vehicles
to sustain
vehicles (EVs)
for
(EVs) are
the growing
foreseeable
arecan growing in popularity.
in future.
popularity. De- tion
sation
in which aa time-varying
time-varying
problem can be
can
the model’s
system.
be system.
classified
classified
structure
ECN
ECN as an
isas
anbattery
battery
fixed
model parametri-
identification
model
identification
while
parametri-
unknown
problem
problem pa-
This
This trend
velopment
trend is
is likely
energy
likely to
to sustain
storage
sustain for
systems
for the
the foreseeable
be
foreseeable future.
considered
future. as De-
the
De- sation
in which problemthe can
model’s be classified
structure as
is an
fixed identification
while unknown problem pa-
velopment
This trend of
is energy
likely storage
to sustain systems
for the can be considered
foreseeable future. as the
De- sation
in which
rameters problemthe
are can
model’s
determined be classified
structure
using as
is an
fixed
measured identification
while
data. unknown
The problem
goal ispa-
to
velopment
heart
velopment of
of vehicle
of energy
energy storage
electrification
storage systems
process.
systems can
can be
Battery
be considered
modelling
considered as
as the
is
thea in which are
rameters the determined
model’s structure
structure
using is fixed
fixed while
measured while
data. unknown
The goal ispa-
to
heart of
velopment vehicle
of electrification
energy storage process.
systems canBattery
be modelling
considered as is
thea in which
rameters
find a model the
are model’s
determined
that its using
output has is
measured
the least data.
deviation unknown
The goal
from ispa-
to
the
heart
critical
heart of
of vehicle
part
vehicleof thiselectrification
electrification process.
technology development.
process. Battery
Battery Theymodelling
are a varie-
modelling is
is a
a rameters
find a modelare determined
determined
that its using
output has measured
the least data.
deviationThe goal
goal
from isthe
to
critical
heart part
ofpart
vehicleof this technology
electrification development.
process. Battery They are
modellinga varie-
is a find rameters
measureda model are
data.that its output
In this using
case, the measured
hasbattery
the least data. The
deviation from
identification problemis to
the
critical
ty of battery
critical part of of this
model
ofmodel technology
types
this technology
technology in thedevelopment.
literature
development. They
which
Theycan are
can a
be
are be varie-
cate- find
measureda model data. that
In its
this output
case, has
the the
battery least deviation
identification from
problem the
ty of battery
critical part this types in the literature
development. which
They are aa varie-
varie-
cate- find
elec- measured
is ana “optimisation
model data.that its output
In this case, the hasbattery
the least deviation
identification from
problem the
ty
ty of
gorized
of
gorized
battery
battery
into
model
into three
model
three
types
groups:
types
groups:
in
(i)the
in
(i)the literature
mathematical
literature
mathematical
which
models,
which
models,
can
can be
(ii)
be
(ii)
cate-
cate-
elec- measured
is
measured data. In
an “optimisation
“optimisation
data. thisproblem”
In this case, the
problem”
case, the inin
which
battery
which
battery
the parameters
identification
the parameters
parameters
identification problem
problem
are
are
ty of
gorizedbattery
trochemical into model
three
models, types
groups:
and in
(i)the literature
mathematical
(iii) electrical which
models,can
equivalent be
(ii) cate-
elec-
circuit is an
optimised to get the problem”
least error in
in which
comparison the to the test data. are
gorized
trochemical into three
models, groups:
and (i) mathematical
(iii) electrical models,
equivalent (ii) elec- is
is an
optimised
an “optimisation
to
“optimisationget the problem”
least error
problem” in
in
in which
comparison
which the
the parameters
to the test
parameters data. are
are
gorized
trochemical
network into three
models,models, groups:
Fotouhiand (i)
et mathematical
(iii)
al. electrical
(2016). models, (ii)circuit
equivalent
Electrochemical elec-
bat- optimised
circuit to get the least error in comparison to the test data.
trochemical
network models,models,Fotouhiand et(iii)
al. electrical
(2016). equivalent
Electrochemical circuit
bat- optimised
optimised to
to get
get the
the least
least error
error in
in comparison
comparison
There are relevant studies in the literature in which the sys- to
to the
the test
test data.
data.
trochemical
network
tery models models,models,
are Fotouhi
the andaccurate
most et(iii)
al. electrical
(2016).
but also equivalent
Electrochemical
the most complex circuit
bat- There are relevant studies in the literature in which the sys-
are relevant
network
tery models
network models,
are
models, Fotouhi
the most
Fotouhi et al.
accurate
et al. (2016).
but
(2016). Electrochemical
also the most
Electrochemical complex bat-
bat- There
tem identification studies in the
techniques areliterature
applied in forwhich
battery thepara-sys-
tery
among models
all are
battery the most
models. accurate
However, butit also
is the
importantmost tocomplex
strike a Thereidentification
tem
There are relevant
are relevant techniques
studies in
studies in the
the literature
areliterature
applied in in which
forwhich
battery thepara-
the sys-
sys-
tery
among
tery models
all
models are
battery
are the
the most
models.
most accurate
However,
accurate but
butit also
is
also the
important
the most
most tocomplex
strike
complex a tem identification
metrisation. Genetic techniques
algorithm are
(GA) applied
is used for battery
battery para-
model
among
balanceallall battery
between models.
model However,
complexity it
and is important
accuracy to
so strike
that a
thea tem tem identification
metrisation.
identification techniques
Genetictechniques
algorithm (GA) are
are by applied for
is used for battery
applied battery
battery para-
model
para-
among
balance
among battery
between
all battery models.
model
models. However,
complexity
However, it
it is
and important
is accuracy
important to
so strike
to that
strike thea metrisation.
identification, Genetic
Hu etalgorithm
al. (2011a), (GA) isconsidering
used for battery model
a complex
balance
models between
can be model
embedded complexity
in and
microprocessors accuracy so
and that
provide the metrisation. Genetic
identification,
metrisation. Genetic
Hu et algorithm
al. (2011a),
algorithm (GA)
(GA) by isconsidering
is used for
used for battery
battery
a model
complex
model
balance
models
balance between
can
between be model
embedded
model complexity
in
complexity and
andet accuracy
microprocessors accuracy so
and that
provide
so In
that the
the identification,
model containing Hu the et al.
ECN (2011a),
model by considering
parameters, SOCa and complex tem-
models
accurate can
resultsbe embedded
in real-time, in microprocessors
Pattipati al. (2011).and provide
other identification,
model containing
identification, Hu the
Hu et al.
et al.
ECN (2011a),
model by
(2011a), by considering
parameters,
considering SOCa and complex tem-
models
accurate
models can
results
can be
be embedded
in real-time,
embedded in
in microprocessors
Pattipati et
microprocessors al. and
(2011).
and provide
In other
provide model
perature containing
at the samethe ECN
time model
and an parameters,
offline SOCa and
optimisation complex tem-
proce-
accurate
words, itresults in
is important real-time,
to have Pattipatimodels et al.that
(2011). In
are accurateother model containing
perature
model containing
at the same the time
ECNand model parameters,
an offline SOC and
optimisation andproce-tem-
accurate
words,
accurate itresults
is
results in
importantreal-time,
to
inunnecessarily
real-time, Pattipati
have
Pattipatimodels et
et al. (2011).
al.that are
(2011). In other
other perature
accurate
In dure
peratureis usedat the
at the fit the
to same
same
ECN
thetimemodel
time
model
and
and toanan
parameters,
offline
experimental
offline
SOC
optimisation
data.
optimisation In aproce-tem-
study
proce-
words,
enough, itandis important
not to have models
complicated. that are
These accurate
reasons dure
peratureis used
atettheto fit the model to experimental data. In a study
words,
enough,
words, it
itandis important
is not to
to have
unnecessarily
important have models
complicated.
models that
that are
are accurate
These reasons
accurate dure
by is used
Brand to same
al. fit thetime
(2014), 31and
model and toan 45 offline
experimental
parameters optimisation
data. In aproce-
are considered study
enough, and
led researchers not unnecessarily
to investigate other complicated.
modelling These
approaches reasons
like dure dureBrand
by is used
is used to fit
et al.
to fit the model
(2014),
the model
31 and to 45
to experimental
parametersdata.
experimental data. In aa study
are considered
In study
enough,
led and
researchers
enough, and not
notto unnecessarily
investigate
unnecessarily other complicated.
modelling
complicated. These
approaches reasons
These network like
reasons by
for Brand
two et
batteryal. (2014),
equivalent 31 and 45
circuit parameters
models are
which considered
are para-
led researchers
electrical circuit to investigate
modelling other
or modelling
equivalent approaches
circuit like by Brand
for
by Brand
two et al.
battery
et al. (2014),
(2014),
equivalent 31 and
31 and 45 parameters
circuit
45 parameters
models are considered
which
are considered
are para-
led researchers
electrical
led researchers circuit to investigate
modelling
to investigate other
or modelling
equivalent approaches
circuit network like for two
metrised battery
using a equivalent
multi-objective circuit models
genetic which
algorithm. are
The para-
main
electrical
(ECN)
electrical circuit
modelling.
circuit Combiningother
modelling
modelling or
or
modelling
lessequivalent
equivalent
approaches
complexitycircuit network
with good
circuit network
like
ac- for for two
two battery
metrised
metrised
battery
using a
equivalent circuit
using a multi-objective
equivalent
multi-objective
circuit models
genetic
models
genetic
which The
algorithm.
which
algorithm.
are para-
are
The
para-
main
main
(ECN)
electrical modelling.
circuit Combining
modelling less
orless complexity with good ac- reason that GA method had been used in those studies was
(ECN)
curacy,
(ECN) modelling.
ECN
modelling. Combining
modelling
Combiningis one ofequivalent
less complexity
the circuit
most common
complexity with
with
network
good
good ac-
battery
ac-
metrisedthat
reason
metrised
reason
using
using
that
GAaa method
GA
multi-objective
multi-objective
method had been
genetic
had beengenetic used in
used
algorithm.
those studies
algorithm.
in those
The main
The
studies
main
was
was
curacy, ECN
(ECN) modelling. modelling
Combiningis one
oneless of the most
complexity common
with good battery
ac- said said
reason to be its greater benefits compared to other methods
curacy,
modelling ECN modelling
approaches is
especially offorthe
EVmost common
application. battery reason to that
be its GAgreater
methodbenefits had been
been used in
compared in to those studies
other methods was
curacy,
modelling
curacy, ECN
ECN modelling
approaches
modelling is
is one
one of
especially offorthe
EV
the most common
common battery
application.
most battery said when toanthat
be GA
its
analytic method
greater
solution had
benefits
does not used
compared
exist and those
to studies
other
when methods
the was
num-
modelling approaches especially for EV application. said
when
said toan be its
analytic greater
solution benefits
does compared
not exist and to other
when methods
the num-
modelling
modelling
ECN battery approaches
approaches especially
models especially
are often for for EV application.
EV application.
parametrised using experi- when ber
whenofoftoan be
unknown
anunknown
its greater
analytic solution
parameters
analytic parameters
benefits
solution does
does
does
compared
arenot
not
existBrand
large,
existBrand
and to when
and when
other
when
methods
et al.the num-
(2014).
the(2014).
num-
ECN battery models are often parametrised using experi- ber
when an analytic solution arenot large,
exist and et al.the num-
ECN
mental battery
data. models
Depending are
on often
the parametrised
application, using
battery experi-
parametri- ber of unknown
Particle swarm parameters are
optimisation (PSO) large,is Brand et al.
employed as (2014).
another
ECN battery models are often parametrised using experi- ber of
Particle unknown
swarm parameters
optimisation are
(PSO) large,is Brand
employed et al.
as (2014).
another
mental
ECN
mental
sation
data. Depending
battery
isdata. modelsoffline
Depending
performed
on
are
on the
often
the application,
parametrised
application,
or online.
battery
battery
In online experi- ber
parametri-
using
parametri-
applications,
of unknown
Particle
optimisation swarm parameters
optimisation
algorithm are
(PSO)
to identify large, Brand
is employed
battery et al. (2014).
as another
parameters from
mental
sation data.
is Depending
performed on
offline the
or application,
online. In battery
online parametri-
applications, Particle
optimisation
Particle swarm
swarm optimisation
algorithm
optimisation (PSO) battery
to identify
(PSO) is employed
is employed as another
parameters
as another
from
mental
sation
speed data.
is
of Depending
performed
the on
offline
parametrisation the
or application,
online.
process In
is battery
online
crucial. parametri-
applications,
An example optimisation
measured test algorithm
data for 12 to identify
different battery
ECN model parameters
structures, from
Hu
sation is performed offline or online. In online optimisation
measured testalgorithm
data for 12todifferentidentifyECN batterymodel parameters
structures,from Hu
speed
sation
speed
of such
of
of the
the parametrisation
isapplications
performed offline
parametrisation
is onlineorprocess
online.
process
battery
is
In
is online applications,
crucial.
crucial. An example
applications,
state-of-chargeAn example
(SOC)
optimisation
measured
et al. (2012). testalgorithm
data for 12todifferentidentify battery
ECN model parameters
structures,from Hu
speed
of such of the parametrisation process is crucial. An example measured
et al. (2012).
measured test data for 12 different ECN
test data for 12 different ECN model structures, Hu model structures, Hu
speed
of such ofapplications
estimation the
applications
as illustrated
is
parametrisation
is online
online
in Fig.
battery
process
battery
1. In isstate-of-charge
this crucial.
state-of-charge
concept, An (SOC)
example
(SOC)
parameters et al. (2012).
of such et al. (2012).
such applications
estimation
of battery
estimation
of
as
as
model
illustrated
applications are is
illustrated
is online
in Fig.
online
in Fig.
estimated
battery
1. In
battery
1. In
online
this
this
state-of-charge
concept,
state-of-charge
in concept,
order to be
(SOC)
parameters
(SOC)
parameters
used by In
et
In al. (2012).
almost all the previous studies, one or more battery models
In almost all the previous studies, one or more battery models
estimation
of battery
estimation as
model illustrated
are
as illustrated in Fig.
estimated
in Fig. 1. 1. In
online this
In thisin concept,
order
concept, to parameters
be used
parameters by are almost all
parameterised the previous
and thenstudies,the model’s one oraccuracy
more battery models
is discussed
of battery model are estimated online in order to be used by In almost
are
In almost all the
parameterised
all the previous
previous studies,
and thenstudies,the model’s one or
one oraccuracy
more battery
more battery models
is discussed
models
of battery model are estimated online in order to be used by are parameterised
of battery model are estimated online in order to be used by are parameterised and then the model’s accuracy is discussed and then the model’s accuracy is discussed
are parameterised and then the model’s accuracy is discussed
Copyright © 2016 IFAC 50
2405-8963 ©
Copyright © 2016, IFAC (International Federation of Automatic Control)
2016 IFAC 50 Hosting by Elsevier Ltd. All rights reserved.
Copyright
Peer review©under
2016 responsibility
IFAC of International Federation of 50 Control.
Automatic
Copyright ©
Copyright © 2016
2016 IFAC
IFAC 50
50
10.1016/j.ifacol.2016.08.008
 IFAC AAC 2016
June 19-23, 2016. Norrköping, Sweden Abbas Fotouhi et al. / IFAC-PapersOnLine 49-11 (2016) 048–053 49
without focus on the parametrisation time. However, the
speed of the battery model identification process can be cru-
cial as well, particularly in online applications. This topic has
been touched by Hu et al. (2011b) where a control-oriented
approach was used for battery model identification using sub-
space method. Based on the previous results in the literature,
it is clear that the battery model parametrisation’s accuracy
and speed depend on the model’s structure and fitting algo-
rithm both and a trade-off is needed between accuracy and
simplicity. The main contribution of this study which distin-
guishes it from the previous works is an “application-
oriented” view on this trade-off. In this study, new discus-
sions are presented which prove that the final target (the final
application) of the battery parametrisation problem can sig-
nificantly affect the choice of fitting algorithm or model
structure. This application-oriented point of view helps to
prevent any unnecessary complexity while achieving the
specified targets as simply as possible.
For this purpose, the battery model parametrisation problem
is analysed using a different visual approach firstly by plot-
ting battery parametrisation surfaces. Although the surfaces
seem simple, they contain quite useful information which is
discussed and utilised in this study. Three standard fitting
algorithms are used and analysed including gradient descent
(GD), genetic algorithm (GA) and prediction error minimisa-
tion (PEM). There are many other algorithms in the literature
and this paper does not aim at reviewing them. GA and PEM
are selected just because they are two standard techniques in
the literature. The GD algorithm is selected to demonstrate
that simpler algorithms may also be applicable for battery
parametrisation. For each level of complexity, the algorithms’
performance is investigated using experimental data from a
small NiMH battery pack. The results of this study are not
limited to NiMH battery chemistry and the proposed contri-
bution can be utilized in other applications as well.
Fig. 1. Online battery state estimation based on parameter
identification
2. BATTERY MODEL IDENTIFICATION
Generally, a system identification procedure consists of three
main parts, Ljung (1987), (i) experiment design, (ii) model
structure selection, and (iii) fitness criterion selection. The
same parts are considered for battery model identification in
this study. Different model structures and data fitting algo-
rithms are used. The algorithms’ computational effort and
precision have been assessed using experimental data for
different model structures.
51
2.1 Battery experiments
As a case study, a six-cell pack of NiMH batteries was tested
using a low-cost test bench that was proposed by Propp et al.
(2015). The NiMH battery pack was selected due to its sim-
ple and save handling as well as its convenient output volt-
age. Specifications of the battery pack are listed in Table 1.
The experiment was conducted at 25°C by applying consecu-
tive discharge current pulses to the battery and measuring the
battery’s terminal voltage. Data is saved in time domain with
a sampling rate of one second. Fig. 2 illustrates the battery
measurements during an experiment. The test started from
fully charged state (8.5 V) and continued until the terminal
voltage dropped below the cut-off voltage (6 V) which means
depleted charge state. The discharge rate is 1C that is 2.4A
and length of each pulse is 40 seconds with a relaxation time
of 60 second in between.
Table 1. NiMH battery pack specifications
Parameter Value
Rated capacity per cell 2400 mAh
No. of cells 6
Rated voltage 7.2 V
Full-Charged voltage 8.5 V
Cut-off voltage 6V
Fig. 2. Measurements during a discharge test by applying
current (input) and measuring terminal voltage (output)
2.2 Battery model structures
Equivalent circuit network (ECN) battery model structures
are used in this study. The ECN battery models are construct-
ed by putting resistors, capacitors and voltage sources in a
circuit. The simplest form of an ECN battery model is inter-
nal resistance model (R model). The R model includes an
ideal voltage source ( VOC ) and a resistance ( RO ) as depicted
in Fig. 3(a) in which Vt is the battery terminal voltage and I L
is the load current. Adding one RC network to the R model
increases its accuracy by considering the battery polarisation
characteristics as discussed by Salameh et al. (1992). This
model, called Thevenin Model (1RC model), is illustrated in
Fig. 3(b) in which Vt is cell’s terminal voltage, VOC is open
circuit voltage (OCV), RO is internal resistance, RP and CP
are equivalent polarisation resistance and capacitance respec-
IFAC AAC 2016
50
June 19-23, 2016. Norrköping, Sweden Abbas Fotouhi et al. / IFAC-PapersOnLine 49-11 (2016) 048–053
tively. Adding more RC networks to the battery model (such
as 2RC in Fig. 3(c)) may improve its accuracy but it increases
the complexity too. Other elements can be added to the model
as well to consider the hysteresis effect for example. As men-
tioned before, this study is not focused on modelling accura-
cy discussion as it has been addressed in a number of previ-
ous studies. However, this study aims at running the system
identification algorithms at different levels of complexity
which depends on the number of unknown parameters. This
is done here by adding more RC networks to battery model.
2.3 Battery identification algorithms
Three identification algorithms are investigated here which
are gradient descent (GD), Snyman (2005), genetic algorithm
(GA), Goldberg (1989), and prediction error minimisation
(PEM), Ljung (1987). Explaining these standard algorithms
is out of the framework of this study and the readers can refer
to the above mentioned references. It should be noted that
there are other algorithms for battery parametrisation in the
literature as well. GA and PEM algorithms are selected as
two standard techniques in the literature which most of the
readers are familiar with them. GD algorithm is selected
based on the results of this study which demonstrate that
simpler algorithms (like GD in comparison with GA and
PEM) may also be applicable for battery parametrisation re-
garding its application. This topic will be discussed in more
details in the following sections.
In order to investigate both the convergence speed and the
accuracy of the battery parametrisation algorithms, different
versions of one algorithm are also considered. For example,
GD’s speed increases by making step size bigger however
this leads to less accuracy. On the other hand, GA accuracy
can be improved by increasing the population size (PS)
and/or number of generations (NoG); however this leads to
less convergence speed. In this study, two versions of GD
algorithm are used; one is more precise (GD1) whereas the
other is faster (GD2). Similarly for GA, two versions are test-
ed; in the more precise version (GA1), PS and NoG are 30
and 35 respectively for the R model whereas in the faster
version (GA2), the parameters are 10 and 15 respectively. For
the 1RC model, more PS is considered; 50 in GA1 and 20 in
GA2. For PEM algorithm, just one version is used.
Generally, a system identification problem can be considered
as an optimisation problem where we are trying to find the
optimum values of a model’s parameters to minimise an error
function. It should be noted that here we are talking about
predefined model structures with unknown parameters. So,
the parameter vector (  ) is determined so that the prediction
error (  ) is minimised, defined as follows:
 (tk
, ) y(tk )  yˆ (tk tk 1; )
where y (tk ) is the measurement data at time k and
yˆ(tk tk 1; ) is the model’s prediction at time k using the
parameters  . In this case, the battery model’s parameters
are optimised so that the least difference between the meas-
ured terminal voltage and the model’s output is achieved. For
this purpose, a fitness function, i.e. the root mean square error
(RMSE) function, is used as follows:
1
1 N
2

2
RMSE    (tk , )  (2)
N k 1 
The battery parameters are not fixed and the optimisation
solution depends on SOC, temperature, etc. Because the bat-
tery identification problem has a time-varying solution, the
identification process should be repeated many times using a
short history of measurements. The current and voltage
measurements are used as the inputs of the optimisation prob-
lem. The identification horizon can be a “time window” or
“SOC window” where the latter is used in this study.
Fig. 3. Equivalent circuit battery models
3. IDENTIFICATION RESULTS
At first, a visual analysis is proposed in this section in order
to get a better understanding about the battery model identifi-
cation problem. RMSE values are obtained as a function of
the unknown battery parameters. Such surfaces are plotted for
R model as depicted in Fig. 4. The model has two unknown
parameters, VOC and RO , which should be obtained during
the optimisation process. The surfaces, shown in Fig. 4, are
obtained by using the measurement history in the past 2%
SOC. For example, for the plot which is obtained at 50%
SOC, the measurements from 48% to 50% SOC are used.
There is a point on each surface at which RMSE has the least
value which is the optimal solution. It should be noted that
the optimum point of the identification surface moves with
regard to SOC however, it keeps its general shape as a
smooth V-shape surface. Looking simple, the battery identifi-
cation surfaces contain significant information about this
(1) particular optimisation problem which has not been discussed
in the literature. These results demonstrate that even a simple
optimisation algorithm like GD might be applicable for bat-
tery parametrisation. Another outcome of these surface plots
is the idea of developing an analytical solution for the battery
parametrisation problem when R model is enough for a speci-
fied application.
52
 IFAC AAC 2016
June 19-23, 2016. Norrköping, Sweden Abbas Fotouhi et al. / IFAC-PapersOnLine 49-11 (2016) 048–053 51

the more accurate model or keeping the faster one. To inves-

tigate the effect of model structure separately, the identifica-
tion results are obtained using a fixed algorithm (i.e. PEM)
and different model structures (i.e. R, 1RC and 2RC models)
as stated in Table 3. RMSE values are quite similar for 1RC
and 2RC and this result is also demonstrated for a short part
of the test in Fig. 8 where the predicted terminal voltage val-
ues are compared vs. experimental data. In other words, there
is no remarkable improvement in voltage prediction by add-
ing the second RC network in this case. Further discussions
about the results are presented in the next section.

Fig. 4. Battery model identification surface (R model) at (a) Fig. 5. Identified values of battery OCV and resistance vs.
10%, (b) 50% and (c) 90% SOC SOC using GA, PEM and GD algorithms

The minimum points of the above mentioned surfaces are

obtained at different SOC levels using GD, GA and PEM
algorithms. Each optimisation solution point contains two
values corresponding to OCV and ohmic resistance. The two
parameters are obtained every 2% SOC using the three identi-
fication algorithms as illustrated in Fig. 5. The parameters are
obtained by applying the algorithms to fit the R model to ex-
perimental data. The fitting quality is measured by RMSE
criterion, introduced in equation (2). The RMSE values and
identification time are compared for different algorithms in
Fig. 6 and Table 2. Based on the results, all the algorithms are
able to get the minimum RMSE for R model however, the
identification time changes using different algorithms.
The results that are presented in Fig. 6 are not valid for other
model structures. This is a part of the main discussion of this
study called “application-oriented” trade-offs between accu-
racy and simplicity. Regarding the application, the simplest
model structure should be selected first and then a suitable
algorithm should be selected after doing the trade-offs. In
order to investigate the effect of the battery model structure, Fig. 6. RMSE and identification time using different algo-
different structures are assessed in this study too. Fig. 7 de- rithms to parametrise R model
picts RMSE and identification time for 1RC model. It is clear
that the trade-offs are different in this case comparing to the
case of R model. The fitting error has decreased by adding
one RC network to the R model. On the other hand, the iden-
tification time has increased by adding more complexity to
the model. So, it is important to see if it is beneficial to use

53
IFAC AAC 2016
52
June 19-23, 2016. Norrköping, Sweden Abbas Fotouhi et al. / IFAC-PapersOnLine 49-11 (2016) 048–053

4. RESULTS ANALYSIS AND DISCUSSION

The results of this study can be classified into three main
parts which are discussed in the followings.

4.1 Study of the battery model parametrisation problem using

a new visual approach
At the first stage of this study, the battery model parametrisa-
tion problem was handled using a new visual approach to get
a better understanding of the problem. Battery model identifi-
cation surfaces (Fig. 4) are obtained which contain useful
insight to the problem. The simplest model structure (R mod-
el) was used to visualize the optimisation problem in two
dimensions. This new presentation of the battery model par-
ametrisation problem, which is done here for the first time,
led us to select simpler algorithms like GD due to the
smoothness of the surfaces. GD algorithm is not applicable
for problems that have lots of local minima however; our
results demonstrate that it can be used for the battery para-
Fig. 7. RMSE and identification time using different algo- metrisation problem. In addition to the use of GD algorithm,
rithms to parametrise 1RC model the idea of developing an analytical solution is also originated
from this visual analysis which can be established in future
studies. As discussed before, the outcomes of such an analyt-
ical solution would be battery OCV and ohmic resistance. It
depends on the application that these two parameters are suf-
ficient or more complexity is needed. This is part of the con-
tribution of this study called application-oriented trade-off
between accuracy and simplicity.
The identification surfaces are obtained at different SOC lev-
els (Fig. 4(a) to (c)). It is demonstrated how the optimum
point of the surface moves with regard to the battery SOC. It
is also demonstrated that the V-shape and smoothness of the
surface remains consistent. Although, other factors like tem-
perature and state-of-health (SOH) are not investigated in this
Fig. 8. Battery terminal voltage prediction vs. experimental study but does not affect our general results and conclusions.
data for 1RC and 2RC models using PEM algorithm The reason is that here we just need to show that the model is
time-varying regardless of the cause of the change.
Table 2. Identification error and time using different algo-
rithms and model structures
4.2 Study of different algorithms for battery parametrisation
Average identification The second part of the results was obtained by comparing the
Average RMSE (mV)
time (s) identification algorithms. Five identification codes are pre-
Method
1RC 1RC pared based on three standard algorithms to assess the trade-
R Model R Model
Model Model off between accuracy and convergence speed. It is obtained
GD1 31.49 28.55 2.19 9.38 from Fig. 5 and Table 2 that the R model’s parameters can be
GD2 31.52 31.20 0.86 1.81 identified using all the three algorithms (GD1, GA1 and
GA1 31.49 26.85 8.48 15.17 PEM) and there is no difference between their accuracy be-
GA2 35.35 29.94 1.30 3.50 cause of the simplicity of the model’s structure. On the other
PEM 31.49 25.70 0.34 1.27 hand, the identification time varies where PEM is the fastest
and then GD and GA. It was also concluded that GA and GD
Table 3. Effect of adding RC networks on battery model algorithms are easily adjustable (by changing their internal
identification error and time parameters like step size or population size) which makes the
compromise easier by trying different versions of them (e.g.
Mean average Max. GD1 and GD2).
Max. Id
Model Method RMSE Id time RMSE
time (s)
(mV) (s) (mV) Based on the results, PEM method is proofed as a suitable
R PEM 31.49 0.34 63.11 0.45 algorithm for battery model identification which has both
1RC PEM 25.70 1.27 44.63 2.22 speed and accuracy at the same time. However, another im-
2RC PEM 25.23 3.01 40.80 3.93 portant criterion that should be considered for battery identi-

54
 IFAC AAC 2016
June 19-23, 2016. Norrköping, Sweden Abbas Fotouhi et al. / IFAC-PapersOnLine 49-11 (2016) 048–053 53
fication algorithm selection is the simplicity of programming
and being easily embeddable. Regarding this aspect, GD al-
gorithm is the simplest one that can be easily programmed
and embedded comparing to PEM and GA. This advantage is
more important for real-time application of the algorithm on
less powerful electronic boards. This also confirms im-
portance of the contribution of this study, which is applica-
tion-oriented trade-off between accuracy and simplicity.
4.3 Study of the trade-off between accuracy and simplicity of
the battery model structures
In addition to the identification algorithm selection, the bat-
tery model structure can also be selected using an applica-
tion-oriented approach. In order to address this, we have fo-
cused on this question that how the identification algorithms
can handle an additional complexity to the model? Here, the
battery model’s complexity is added gradually by putting
more RC networks in the model which increases the number
of unknown parameters from the identification point of view.
Considering the results presented in Fig. 6 and Fig. 7, it is
concluded that GD algorithm’s speed decreases significantly
by adding model’s complexity while keeping the accuracy in
a comparative level. In GD algorithm, the identification time
that is needed for a battery model with k unknown parame-
ters is proportional to 2k . For example, the time needed to
identify 1RC model (which has 4 parameters) is 4 times more
than R model (which has 2 parameters), and the time needed
to identify 2RC model (which has 6 parameters) is 16 times
more than R model and so on. Therefore, GD method is not
suggested for very complex models however it is a good
choice when a simple and easily embeddable identification
algorithm is needed for online applications.
On the other hand, GA algorithm may not be successful in a
short period of time however the complexity can be handled
if there is no time limitation such as in offline applications. In
this case, GA algorithm outperforms GD because of its addi-
tional capabilities like jumping out of local minima. So, GA
method is suggested to be used for very complex models
when time is not an important factor. In addition, GA algo-
rithm is difficult to be embedded because of its programming
complexity. Referring to PEM method’s results in Table 3
and Fig. 8, there is an improvement in RMSE by adding the
first RC network. However, this achievement is obtained by
increasing the identification time to four times more. Adding
the second RC network causes a small improvement of error
while the identification time approximately doubles. So, it is
concluded that 2RC battery model structure is not suggested
in this case. All in all, PEM method is suggested when bat-
tery model’s complexity is needed in online applications
however; its embeddable program is not as simple as GDs.
5. CONCLUSION
In this study, a new approach to the battery parametrisation
problem was presented which is called application-oriented
trade-off between accuracy and simplicity. The main ad-
vantage of this approach is avoiding unnecessary complexi-
ties by using the most efficient model structure and para-
metrisation algorithm in each case. The results of this study
55
are mainly usable in predictive models rather than descriptive
models. A good example of the predictive modelling applica-
tion is real-time estimation of battery SOC and/or SOH. The
choice of the parametrisation algorithm and the model struc-
ture depends on the required accuracy, required speed, com-
putational effort and hardware limitations, etc. For the pre-
sented case, a NiMH battery, SOC estimation is possible just
by using OCV which is available using the simplest model
structure that is R model. Both PEM and GD algorithms are
suggested in this case depending on the computational effort
and hardware limitations. Considering another battery type,
Lithium-Sulfur for example, OCV is not enough for SOC
estimation and other parameters are required as well. Same
discussions can be presented for battery SOH estimation
which is another application for battery model parametrisa-
tion.
REFERENCES
Brand, J., Zhang, Z., Agarwal, R.K. (2014). Extraction of
battery parameters of the equivalent circuit model using
a multi-objective genetic algorithm. Journal of Power
Sources, 247, 729-737.
Fotouhi, A., Propp, K., Auger, D.J. (2015). Electric Vehicle
Battery Model Identification and State of Charge Esti-
mation in Real World Driving Cycles, 7th Computer
Science and Electronic Engineering Conference, UK.
Fotouhi, A., Auger, D.J., Propp, K., Longo, S., Wild M.
(2016). A review on electric vehicle battery modelling:
From Lithium-ion toward Lithium–Sulphur, Renewable
and Sustainable Energy Reviews, 56, 1008-1021.
Goldberg, D.E. (1989). Genetic Algorithms in Search, Opti-
mization & Machine Learning, Addison-Wesley.
Hu, X., Li, S., Peng, H. (2012). A comparative study of
equivalent circuit models for Li-ion batteries. Journal of
Power Sources, 198, 359-367.
Hu, Y., Yurkovich, S., Guezennec, Y., Yurkovich, B.J.
(2011a). Electro-thermal battery model identification for
automotive applications. Journal of Power Sources, 196,
449-457.
Hu, Y., Yurkovich, S. (2011b). Linear parameter varying
battery model identification using subspace methods.
Journal of Power Sources, 196, 2913-2923.
Ljung, L. (1987). System Identification - Theory for the User,
Prentice Hall, New York.
Pattipati, B., Sankavaram, C., Pattipati, K.R. (2011). System
Identification and Estimation Framework for Pivotal
Automotive Battery Management System Characteris-
tics. IEEE Trans. Syst. Man Cybern. C: Appl. Rev.,
41(6), 869-884.
Propp, K., Fotouhi, A., Auger, D.J. (2015). Low-Cost Pro-
grammable Battery Dischargers and Application in Bat-
tery Model Identification, 7th Computer Science and
Electronic Engineering Conference, UK.
Salameh, Z.M., Casacca, M.A., Lynch, W.A. (1992). A
mathematical model for lead-acid batteries. IEEE
Transactions on Energy Conversion, 7(1), 93-98.
Snyman, J.A. (2005). Practical Mathematical Optimization:
An Introduction to Basic Optimization Theory and Clas-
sical and New Gradient-Based Algorithms, Springer
Publishing.