Energy 60 (2013) 35e43

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

Capacity fade and aging models for electric batteries and optimal
charging strategy for electric vehicles
I.J. Fernández, C.F. Calvillo*, A. Sánchez-Miralles, J. Boal
Institute for Research in Technology (IIT), ICAI School of Engineering, Comillas Pontifical University, Santa Cruz de Marcenado 26, 28015 Madrid, Spain

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

Article history: Due to the increasing concern for energy efficient storage devices, extensive research is being done on
Received 29 August 2012 the mechanisms which are involved in the battery degradation process. Optimality in electric vehicle
Received in revised form energy management has been traditionally pursued from the perspective of efficient grid operation, but
9 July 2013
barely considering battery degradation in the process. The objective is to formulate a battery degradation
Accepted 31 July 2013
model which will yield the degradation suffered by a battery as a function of its use, allowing both the
Available online 7 September 2013
comparison between different energy management strategies as well as including the model in linear
optimization algorithms.
Keywords:
Battery degradation models
Nearly three-hundred tests were carried out on Li-ion battery cells which were subjected to different
Capacity fade depth-of discharges, currents and temperatures. Results are used to identify the variables which cate-
Self-organizing maps gorize best, using self-organizing maps, the degradation suffered by the cells. A model is proposed which
Electric vehicles yields the rate of degradation of the battery as a function of both temperature and depth-of-discharge.
Energy management This model is then linearized and used in an electric vehicle energy management optimization problem,
where the degradation suffered by the battery due to a controlled charge is minimized. Finally, the re-
sults are compared with other charging methods.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction The degradation of the electric battery compares to the aging of

The electric vehicle appears as a promising solution to effec-
tively reduce GHG (greenhouse gas) emissions and other environ-
mental problems. Smith [1] analyzed the effect of the electrification
of individual vehicles, resulting in considerable and immediate
reductions in GHG emissions for urban driving cycles. Furthermore,
many authors remark the importance of the energy storage ca-
pacity in the electric vehicle for higher penetration of renewable
sources [2] and to improve grid reliability [3,4]. Two critical vari-
ables in battery operation, which are rarely observed with the
required detail, are the SOC (State of Charge) and the battery
degradation. Sun et al. [5] remark the importance of the former,
stating that an imprecise SOC estimation can easily cause under- or
over-charging situations, affecting the power-output capability and
the lifetime of the battery pack. The latter appears in all battery-
based systems; thus, an accurate assessment of degradation is
needed to develop optimal operation schemes and feasibility
analysis [6].
* Corresponding author.
E-mail addresses: ismael.fernandez@iit.upcomillas.es (I.J. Fernández),
christian.calvillo@iit.upcomillas.es (C.F. Calvillo), alvaro@upcomillas.es
(A. Sánchez-Miralles), jaime.boal@iit.upcomillas.es (J. Boal).
a vehicle fuel tank. A new fuel tank can hold its full capacity and
may easily provide the engine with all the energy stored at a
desired rate. As the tank is refilled, the bottom of the tank may
accumulate sediments which slowly decrease its capacity. These
sediments could clog the fuel tank filters which would limit the rate
at which fuel could be extracted. To make matters worse, the fuel
tank may hypothetically develop small holes, like Somogye’s
bucket [7], through which fuel would be lost. In the same manner,
the electric battery loses capacity with use throughout its lifetime,
decreasing the total amount of energy it can store and deliver. As
the battery degrades, the internal resistance increases resulting in a
difficult extraction of the stored energy, in addition to the loss of
part of this energy in the form of heat during charging and dis-
charging. Lastly, batteries self-discharge, causing them to slowly
lose energy despite that the battery is not being used.
Due to the increasing concern for energy efficient storage de-
vices, for use in alternative modes of transportation and for use in
ancillary grid services, extensive research is being done on the
mechanisms that are involved in the battery degradation process.
Understanding why the battery degrades helps developers to focus
efforts where the battery suffers most. Although there is much
literature on this area, there is little on the quantification of the
battery degradation as a function of its use, in order to estimate

0360-5442/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.energy.2013.07.068
36 I.J. Fernández et al. / Energy 60 (2013) 35e43
Nomenclature
Sets
p period
s discrete charging power index
seg piecewise function segment
Parameters
pBattSOH battery state-of-health
pDiscPowValuess available discrete power charging level s
[kW]
pSOCMin minimum admissible state-of-charge [p.u.]
pEff overall charge/discharge efficiency [p.u.]
pAvailabilityp indicates whether the vehicles is plugged in
during period p or not 0.1
pSOCini state-of-charge of the electric vehicle at time t ¼ 0
[kWh]
pConsumptionp energy consumption of the electric vehicle
during period p [kWh]
pTemperaturep temperature forecast for period p [ C]
Variables1
vPowerCommandp charging command for the vehicle during
period p [kW]
vDiscPowIndexp,s discrete power level s chosen for period p
[kW]
vSOCp state-of-charge of the vehicle at the end of period p
[p.u.]
vDODp depth-of-discharge of the vehicle at the end of
period p [p.u.]
vBatteryDegradationp battery degradation for the vehicle
during period p [p.u.]
both the cycle life and calendar life of the battery. Equally important
is the understanding of how much it degrades. It is crucial for any
kind of energy storage, for electric vehicle manufacturers and for
vehicle-to-grid implementations, to be able to estimate the battery
degradation as a function of specific charge/discharge regimes and
conditions.
According to Marano et al. [6], there are two approaches to
battery life estimation models: performance-based models and
weighted Ah-throughput models. The former simulates the change in
battery parameters such as capacity, voltage or current, and de-
termines the end-of-life of the battery when a certain threshold has
been reached. Under these models, we have included the electro-
chemical models [8e10], equivalent circuit models [7,11e13] and
some models based on neural networks [7,11,14]. Performance-
based models provide a very detailed model of the battery, with
extensive information and results on local conditions and perfor-
mance [6]. They are mainly used to optimize the physical design
aspects of batteries, characterize the fundamental mechanisms of
power generation and relate battery design parameters with
macroscopic and microscopic information [13]. These models can
be very complete; however, these models are complex to build and
understand, in addition to time consuming as they require
knowledge of the chemical and physical interactions within the
battery as well as coupled time-variant spatial partial differential
1
The auxiliary variables used to model the battery degradation have not been
cited in this section for the sake of brevity. Notwithstanding, they are introduced
and explained in detail in following subsections.
equations [13,6]. Electrochemical models predict the effects of
some parameters such as time and temperature accurately, but the
different models found do not agree well with cycle life test data
[15]. Equivalent circuit models, although simple, are mostly used
for performance rather than life estimation purposes.
The second approach that Marano et al. [6] mentioned is the
weighted Ah-throughput models. These models link the end-of-life
with a parameter which can be easily measured such as Ah-
throughput, number of cycles or time since manufacture. These
models are based on the assumption that, under some standard
conditions such as current, temperature or DOD (depth-of-
discharge), a battery can achieve an overall Ah-throughput before
its end-of-life is reached [6]. Drouilhet et al. [16] defines three
premises related to this approach. The first premise says that each
cell has a finite life as measured by the sum of the effective ampere-
hours throughput during its useful life. The second premise states
that the actual charge life of the cell, not just its cycle life, is a
function of the depth of discharge at which it is cycled. Lastly, the
third premise indicates that the charge life of the cell will be
decreased whenever the cell is discharged at a rate faster than the
rated rate.
Traditionally, optimality in electric vehicle energy management
has been pursued from the perspective of efficient grid operation,
but barely considering battery degradation in the process. For
instance, the economic feasibility for electric vehicles is studied in
Refs. [17,18], achieving interesting results; however, battery
degradation is disregarded. On the other hand, penalty for exces-
sive or incorrect use of the batteries can be seen in several articles
which penalize high depth-of-discharge or the amount of Ah-
throughput [19e21], some taking the naive approach of limiting
the number of cycles the battery may go through daily for the sake
of simplicity [22,23] or by just keeping the battery state of charge
between what is considered acceptable levels [24]. These algo-
rithms do not enter in great detail into the degradation models of
the batteries, only including Ah-throughput or DOD restrictions.
Furthermore, temperature is not considered during optimization,
which is crucial for the correct use of the battery and for cost
reduction in system optimization. The objective of this paper is to
develop a degradation model of a Li-ion battery which can be
applied to optimization problems in order to evaluate, compare and
minimize battery deterioration.
The present paper is divided into two parts, the first part
focused on developing a degradation model of the electric battery
and the second part where the proposed model is used in an
optimization problem, developing an optimal charging strategy.
Section 2 focuses on a first concern: the causes of battery degra-
dation. Before developing a model for battery degradation it is
important to understand the causes why the battery degrades, in
order to experiment and model them appropriately. In Section 3 the
experimental procedure, data collection and interpretation are
described. Then, a degradation model is developed using the data
collected. Section 4 shows an energy management optimization
application for electric vehicles, where an algorithm is proposed to
minimize battery degradation as a function of DOD and tempera-
ture; then, it is compared to other charging methods. The paper
ends with Section 5 where concluding remarks are given.
2. Causes for battery degradation
There are a number of producer agents of degradation and loss
of battery capacity in Li-ion based batteries. Many researchers
agree that the most important causes are lithium deposition, pas-
sive film formation, crack propagation and active material disso-
lution in the electrode [8,9,15,16,25e27]. These processes may be
accelerated with battery misuse, but some are also present during
 I.J. Fernández et al. / Energy 60 (2013) 35e43
normal operation [9]. Battery degradation is complicated and
strongly influenced by charge and discharge conditions; for
instance, higher C-rates, higher maximum voltage level or extreme
temperatures accelerate degradation. Additionally, a battery pack is
as weak as its weakest cell, so as soon as one cell is degraded, it will
drag the whole battery string resulting in a premature failure [16].
Battery misuse accelerates these degrading processes and it is
therefore crucial to understand what is considered an abusive
conduct. In the following subsections abusive conducts which
contribute to battery degradation will be discussed.
2.1. Battery use
SEI (Solid Electrolyte Interface) protects the negative electrode
from aggressive solvents in the electrolyte. This film forms during
the first charge, however it captures some amount of electro-
chemically active lithium, increasing the internal resistance and
decreasing the capacity [15]. According to Danilov and Notten, the
SEI formation is generally accepted as one of the main processes
responsible for battery degradation [28]. Similarly, J. Li et al. state
that under normal charge-discharge regimes, the change in ohmic
resistance comes mainly from the formation of the passive films on
the surface of the electrodes during cycling [9].
2.2. Rate of charge/discharge
Effects of high rates of charge or discharge are briefly explained
in Ref. [16], mentioning that they cause a greater loss in conduc-
tivity between adjacent particles in the active material matrix,
increasing the internal resistance. High rates of charge and
discharge are the main contributor for Lithium-deposition, caused
when the flow of lithium ions is greater than the one which can be
intercalated in the electrode. When this occurs, lithium ions deposit
as metallic lithium which captures active lithium and may even
cause a short-circuit between the electrodes. On the other hand,
Premanand et al. carry out studies on capacity fade for spinel based
Li-Ion cells with different charge currents, achieving interesting
results. Premanand et al. conclude that for all the charge currents
tested, the resistance of both electrodes in the batteries did not vary
significantly [25]. Contrary to the results mentioned, Drouilhet et al.
also cited a reference by H. Wenzl, Understanding Batteries e The
effect of charging and discharging, where there is evidence that at
low and intermediate states of charge, high charge rates are not
harmful for the battery and in fact may increase the battery life [16].
There seems to be little research or documentation on this last
theory and has not been found again in other references.
2.3. Temperature
High temperature in batteries has a double effect on the battery
performance. As the temperature in the battery increases, as
explained by the Arrhenius relationship, the rate of the chemical
reactions involved in charging or discharging will also increase [15].
This increased rate causes the battery’s internal resistance to
decrease, increasing the efficiency and performance of the battery.
On the counterpart, high temperatures also accelerate unwanted
chemical reactions which degrade and reduce the capacity of the
electric battery [29,15]. With higher temperatures, the chemical
reaction of reduction on the surface of the carbon electrode will
occur at a faster rate, causing gas. With this increased pressure and
temperature, the surface film on the electrode can be easily
stretched or damaged. This forms cracks through which lithium
ions and electrolyte can react with the lithiated carbon particles
from the electrode [26], forming more SEI and therefore capturing
more active lithium particles. Similarly, Magazine et al. claim that
37
the temperature may cause the separation of the anode or cathode
material from the electrical collector metal [15].
Temperature control becomes crucial, but sometimes very
difficult to accomplish. Battery packs usually come with many cells
packed close together, and heat dissipation becomes difficult. Ning
et al. report evidence that the difference between the internal
temperature of a single cell and the external cell can be up to 60
Celsius [26]. Magazine et al. compare battery degradation with
physical stresses using the Zhurkov formula, which indicates that
the rate of degradation increases exponentially with the stress
level, which is a function of the temperature [15]. Magazine et al.
give the following statement: “if the aging of a battery, at room
temperature, were to be about ten years, the aging process speeds
up by a factor of two every 10  C.” Similarly, Serrao et al. also
mention that the rate of unwanted chemical reactions will double
as the temperature increases by ten degrees [29]. Magazine et al.
conclude that the battery life can be maintained if the battery
temperature is kept low (<30  C) [15].
2.4. Depth of charge/discharge
A cell which is cycled with a high depth-of-discharge deteriorates
quickly due to a constant stripping and re-depositing of the solid
electrodes [7]. Some references use models based on the concept of
accumulated charge throughput in order to estimate the capacity
loss or battery aging [6]. In other words, these models assume that a
battery has a total amount of energy throughput: with every
discharge the amount of energy can be subtracted from the total
estimated amount obtaining a value proportional to the degradation.
This method explains why a battery will last significantly more cycles
with small DODs rather than with larger DODs [29]. Battery life can
be maintained in an acceptable range for plug-in hybrid electric
vehicles if very deep cycles, (>0.6 DOD) are avoided [15].
2.5. Overcharging
Capacity fade can be caused by a number of different reasons.
According to Ning, while developing a model and defining an
equation for the capacity fade for lithium-ion batteries, there are
three parts to the equation which models the capacity fade of a
battery [26]. The first part is the correction term due to different
rate capability. The second part corresponds to the capacity fade
due to the loss of the carbon electrode and the third is the capacity
fade due to the loss of lithium ions and other reasons. Ning con-
cludes that the capacity fade is mostly caused by the second term:
the increase surface resistance due to the oxidation process. Over-
charging causes irreversible partial decomposition of the LiCoO2
electrode [9], forming Cobalt oxide and Lithium dioxide as can be
seen in (1), resulting in an increase of surface resistance and loss of
active lithium.
Liþ þ e þ LiCoO2 /Li2 O þ CoO (1)
3. Battery model
3.1. Definitions
The variables, definitions and mathematical formulations used
throughout this paper are defined as follows:
 Cycle A cycle is considered finished when the current passes
through zero and changes sign.
 C-rate The current corresponding to the manufacturer’s rated
capacity, in ampere-hours, for a 1 h discharge at 30  C. For
38 I.J. Fernández et al. / Energy 60 (2013) 35e43

Voltage Profile during Static Capacity Test  Degradation The battery degradation, given as a fraction of the
4.6 cell’s capacity which has faded relative to the EOL, is given by
Equation (4).
4.4

4.2 Qt  Qo
BatteryDegradationt ¼ (4)
Qo  0:8Qo
4

3.8
 Ah_Throughput This is the Ah that cell has transferred be-
Voltage (V)

3.6 tween times t1 and t2, given by Equation (5).

3.4 Zt2
Ah Throughputðt1 ; t2 Þ ¼ absðIðtÞÞdt (5)
3.2
t1
3

2.8  State of Charge Also known as SOC, it is the available capacity

in a battery or cell, expressed as a percentage of the SOH.
2.6  Depth-of-Discharge Also know as DOD, it is the percentage of
0 100 200 300 400 500 600
the cell’s SOH removed by discharge or added by charge.
Time (m)
 Effective Throughput Cycle The effective throughput cycle,
Fig. 1. Voltage profile during a static capacity test. Neff, is adapted from Ref. [15]. The effective throughput cycle
accounts for the difference between cycling the battery at
example, if the battery’s rated capacity is 40 Ah, then the C-rate different DODs. Neff is defined by Equation (6).
is 40 A as defined by Ref. [30].
 Static Capacity Also known as the SOH (State of Health), is the
cell total capacity at a given time, measured in Ah and defined Z
1 absðIðtÞÞdt
by Equation (2). Neff ¼ (6)
2 Qo
Z time

Qt ¼ IðtÞdt (2) In other words, a cell cycled at a depth-of-discharge of 100%

fulldischarge would have an effective throughput cycle of one after one charge
and discharge, while a cell cycled at a depth-of-discharge of 10%
would require ten cycles to have Neff ¼ 1.
 The test procedure followed to measure the Static Capacity,
was adapted from Ref. [30] and it is summarized as follows. The
3.2. Experimental procedure
specific cut-off voltages and currents given correspond to
Lithium-ion cells with a LiCoO2 cathode.
Twenty seven lithium-ion battery cells were tested during a
1. The battery cell is brought to room temperature for a period
period of one year at different temperatures, currents and DODs.
of at least 5 h to allow the battery cell to reach equilibrium.
Almost 300 test were carried out, including over 22,000 charge/
2. The battery is fully charged according to manufacturer
discharge cycles. The battery cell used during experimentation was
specifications. This is a CCCV (Constant Current, Constant
a 10 Ah polymer lithium-ion battery cell, model PL-9059156.
Voltage) charge, charging at a constant current (0.6 C) until
Currently in the market there are different sorts of lithium-ion
4.2 V are reached, followed by a constant voltage charge
batteries depending on the chemistry of the cathode. In this case,
until the current declines to 0.05 C.
the lithium-ion battery type chosen has a LiCoO2 cathode. These
3. The battery is allowed to settle for 1 h. This allows for the
types of batteries have the highest energy density [W/kg] and are
chemicals and the temperature to stabilize.
one of the most popular cell types used.
4. The battery is fully discharged at a constant current equal to
Out of the total number of battery cells tested, one-third of the
0.6 C until 2.75 V are reached, following manufacturer
batteries where introduced into a ESPEC temperature chamber
specifications.
(Model SU-220), where the temperature was kept constant at 5  C.
One-third of the total number of battery cells where introduced
The resulting voltage profile during a static capacity test can be
into a Nahita Drying Oven, where the temperature was kept con-
seen in Fig. 1.
stant at 50  C. Lastly, the remaining batteries cells were kept at
room temperature. A BTS (battery test system) from Neware was
 Capacity Fade The capacity fade, given as a percentage, is
used to record the data and execute the tests. At each temperature
defined as the capacity that the cell has irreversibly lost due to
level tested, the batteries were further divided into eight groups,
battery degradation. It is given by Equation (3).
where there were batteries cycling at high, medium and low cur-
rents, (0.9 C, 0.55 C, 0.1 C respectively) and different DODs, ranging
 
Qt from 80% to 10%.
CapacityFadet ¼ 100, 1  (3)
Qo The experimentation consisted on the following steps:

Where Qo is the static capacity of the new cell and Qt is the static
capacity at any given moment t.
 End-of-Life The EOL (end-of-life) of a cell will be considered
reached when the capacity fade has reached 80%.
1. The Static Capacity Test, defined in Section 3.1, was initially
carried out on every battery cell in order to measure the initial
capacity Qo. Although the manufacturer claims that the battery
capacity was initially 10 Ah, this was proven different for every
cell.
 I.J. Fernández et al. / Energy 60 (2013) 35e43 39

2. The temperature chamber and drying oven were both powered Histogram
on and allowed between 5 and 10 h for the temperature to 250
stabilize before starting the tests.
3. A number of cycles were carried out at the chosen DOD and
current. To guarantee a stable average SOC, the battery cell was 200
first charged, using CCCV, to a voltage V or cut-off current A.
When reached, the cell was discharged at the chosen current
rate until the desired DOD had been reached. The cell was then
150
charged once again, using CCCV, to the same voltage V or cur-

Count
rent A. This provided a known starting point for the discharge
to begin. All CC charges were carried out at the chosen test
current rate. 100
4. After a number of cycles the battery cell test was stopped, and
the temperature chamber and drying oven were powered off.
The batteries were allowed between 5 and 10 h to stabilize and 50
reach room temperature. A static capacity test, defined in
Section 3.1, was then carried out to measure the SOH at time t.
5. Steps 2 to 5 were repeated until the EOL had been reached or
0
the testing period had come to an end. 0 1 2 3 4 5 6 7
Degradation / N −6
x 10
The experimental battery degradation results were analyzed to
determine the distribution characteristics of the data. Chi-square Fig. 3. Histogram of resulting degradation per cycle profile.

goodness-of-fit tests were carried out to prove if the experi-

mental results follow a normal distribution. The tests indicate that
the 88.9% of the data sets correspond to a normal distribution with
a 95% confidence; thus, since most of the normality tests accept the
null hypothesis, it can be assumed that the experimental results
follow a normal distribution. Fig. 2 show the equivalent distribu-
tions for all data sets.
3.3. Model design
In order to analyze the relationship between the variables when
classifying the rate of degradation, SOM (self-organizing maps)
were used. These are a type of artificial neural network which is
trained using unsupervised learning. By analyzing the resulting
maps and applying labels afterwards, it is possible to identify a
relationship between the desired output and the inputs.
The first step was to define the intervals which would later
define the labels used in the SOMs. To distribute the labels equally
throughout the test cases, the histogram of the degradation per
cycle (Fig. 3) was analyzed. It can be seen that most of the test cases
fell in a very small region. Further analysis provides proof that a
logarithmic scale should be used to obtain equal number of cases in
each interval. The data was labeled as low-degradation, medium-
degradation, high-degradation and very high degradation accord-
ing to the scale shown in Table 1. The table shows the labels as well
as the resulting number of cases which fell in each category.
For the sake of brevity, only the self-organizing maps for the
significant variables are shown. Nevertheless, a study was made
with all of the variables and different combinations of them, with
the intention of finding a possible relationship that would explain
the different degradations. Using this method of study, several
variables have been discarded, such as total Ah-throughput, initial
degradation before each test, average voltage or average SOC. The
results show that using temperature and depth-of-discharge, the
degradation was very well categorized.
The SOM for the temperature and for the DOD, Fig. 4 and Fig. 5
respectively, clearly show the relationship between temperature
and DOD in categorizing the degradation suffered by the cells. A
high DOD, whether at high or low temperatures, corresponds to a
very high degradation. Similarly, a high temperature corresponds
to a high degradation despite the level of DOD. Moreover, low
temperatures also cause a high degradation on the battery for DODs
above a certain threshold.
After identifying the two most significant variables which
categorize the battery degradation, we introduced the acquired
data points into a curve-fitting software in Matlab with tempera-
ture and DOD as inputs and the rate of battery degradation per cycle
as the output. The resulting curve is shown in Fig. 6. This curve
defines the degradation expected with one charge or discharge
cycle at a given temperature and DOD. The error of the developed
model was calculated for all the experimental data. Table 3 presents

DOD = 10% DOD = 45% DOD = 80%

900 1400 120

800
1200 100
700
1000
600 80
800
Count

500
60
400 600
300 40
400
200
200 20
100

0 0 0
0 0.005 0.01 0.015 0.02 0.025 0.03 0 0.005 0.01 0.015 0.02 0.025 0.03 0 0.005 0.01 0.015 0.02 0.025 0.03
Degradation/N Degradation/N Degradation/N

Fig. 2. Normal distribution of the results, separated by DOD and Temperatures.

40 I.J. Fernández et al. / Energy 60 (2013) 35e43

Table 1
SOM Temperature
Intervals used for SOM labels.

Degradation Label Lower limit Upper limit Cases

Low 1 N 3  105 25% 4 3 3 3 4

Medium 2 3  105 2  104 25%
High 3 2  104 1  103 28%
Very High 4 1  103 þN 22% 4 4 4 4

4 3 1 3 3 3
the separated absolute errors for each temperature and DOD cluster
in the corresponding percentage. It can be seen that batteries
4 3 3 2 3 2 3
behave more regularly at room temperature than at other tem-
peratures; therefore the error is smaller. Similarly, battery degra-
dation at lower DOD has less variation, producing less absolute 2 1
mean error in the model, as can be seen in Fig. 2. The total mean
absolute error of the model is 59%, this high value is caused by the 1 1 1 2 2 1 2 3
nature of the data, which is very close to zero; therefore, even a tiny
variation in the absolute value could produce a large error.

4. Applications 0.1 0.2 0.3 0.4 0.5 0.6

This section is meant to show possible applications of the Fig. 4. Self organizing map for temperature.
developed model. The section evaluates different charging strate-
gies, as well as the possibility of including a penalty for battery
degradation during the management of the state of charge of predicted consumption during period p. This prediction is an
electric vehicles. The section compares the degradation suffered by input to the model and its calculation is out of the scope of this
an unmanaged charge of the electric vehicle, a simple model which paper.
minimizes the sum of the depth-of-discharges that the battery is
subjected to during the day, and a model which includes a penalty vSOCp ¼ vSOCp1 þ pEff ,vPowerCommandp ,dT
for battery degradation as a function of DOD and the temperature.  pConsumptionp cp (8)
The proposed model allows for both the computation of the
degradation suffered using different charging strategies, as well as
pSOCMin  vSOCp  pBattSOH cp (9)
the possibility of including a degradation model in a linear opti-
mization problem. The power commands sent to the vehicles must have a discrete
value due to constructive limitations of the electric vehicle char-
4.1. Introduction gers. In order to model this, the following two equations have been
implemented.
As was mentioned in Section 1, optimality in electric vehicle X
energy management has been traditionally pursued from the vPowerCommandp ¼ pDiscPowValuess ,vDiscPowIndexp;s cp
perspective of efficient grid operation, but barely considering bat- s
tery degradation in the process. The objective of the optimization (10)
algorithm set forth is to minimize the degradation that the battery
will suffer throughout the day, due to the charge and discharge
process. The degradation will be calculated as a function of the
depth-of-discharge and as a function of temperature using the
SOM Depth−of−Discharge
degradation model from Section 3.

4.2. Mathematical formulation 4 3 3 3 4

This section presents the detailed mathematical formulation of 4 4 4 4

the model used to optimally charge an electric vehicle.
4 3 1 3 3 3
4.2.1. Objective function
The objective function (7) minimizes the degradation suffered
4 3 3 2 3 2 3
by the electric battery due to cycling throughout the day.
X
min vBatteryDegradationp (7) 2 1
p

1 1 1 2 2 1 2 3
4.2.2. Constraints
The model must satisfy the following system and operational
constraints.
The state-of-charge (8) must be within limits (9) for every 0.2 0.3 0.4 0.5 0.6 0.7
period p. The SOC is calculated as an inventory, adding to the SOC
at p1, the energy charged during period p and subtracting the Fig. 5. Self organizing map for depth-of-discharge.
 I.J. Fernández et al. / Energy 60 (2013) 35e43 41

Battery degradation per cycle as a function of Temperature and DOD Piecewise linear function of battery degradation model

−3
x 10
0.025 15
0.02
Degradation / N

10

Degradation / N
0.015

0.01
5
0.005

0 0

−0.005
1 −5
60 1
0.5 40 50
40
20 0.5 30
0 0 20
DOD [−] Temperature [ C] 10
0 0
DOD Temperature C
Fig. 6. Resulting curve for battery degradation as a function of temperature and DOD.
Fig. 7. Resulting piece-wise linear curve for battery degradation as a function of
temperature and DOD.
X
vDiscPowIndexp;s ¼ 1 cp (11)
s

vCalcDODUpp þ vCalcDODDnp ¼ 00vSOCAuxp ¼ vSOCAuxp1 cp

4.2.3. Battery degradation
In order to model battery degradation, first it is necessary to (19)
calculate the two inputs to the model: the battery depth-of- Finally, the absolute value of the DOD is calculated as follows.
discharge and the temperature. The temperature is a parameter
to the system, therefore only the DOD has to be calculated. The vSOCStatep ¼ 10vDODp ¼ vSOCAuxp  vSOCAuxp1 cp
following logical conditions have been modeled using linear
(20)
functions.
The first set of implications use state variables to indicate
whether the SOC is increasing (12), decreasing (13), or stable (14).
vSOCStatep ¼ 00  vDODp ¼ vSOCAuxp  vSOCAuxp1 cp
(21)
vSOCp  vSOCp1 > 05vSOCUpp ¼ 1 cp (12)

vSOCp  vSOCp1 < 05vSOCDnp ¼ 1 cp
vSOCp  vSOCp1 ¼ 05vSOCStblp ¼ 1 cp
The cycle is not considered finished until the current has
changed sign, and therefore a stable SOC indicates that it might still
be increasing or decreasing. The variable vSOCStatep is equal to one
when the SOC is increasing and equal to zero when decreasing. The
following two implications define this.
vSOCStblp ¼ 10vSOCStatep ¼ vSOCStatep1
vSOCStblp ¼ 00vSOCStatep ¼ vSOCUpp cp
The last step indicates when the current changes sign and,
consequently, when the DOD must be calculated. The variables
vCalcDODUpp and vCalcDODDnp indicate when the DOD must be
calculated and whether it was increasing or decreasing.
vSOCStatep vSOCStatepþ1 vCalcDODUpp þvCalcDODDnp
¼ 0 cp
The last auxiliary variable maintains the last value of the SOC
between two cycles, and is updated only when the DOD must be
computed according to (17).
crvCalcDODUpp þ vCalcDODDnp ¼ 10vSOCAuxp ¼ vSOCp
(13) 4.2.4. Degradation as a function of temperature and DOD
The piecewise linear functions shown in Fig. 7 are a linear
(14) approximation of the surface in Fig. 6. These were obtained using
the recursive DouglasePeucker line simplification algorithm, which
is used to reduce the number of vertices in a piecewise linear curve
according to a specified tolerance. This tolerance was fine-tuned to
obtain a maximum of four segments in any given piecewise linear
curve. The five piecewise linear curves shown in Fig. 7 have been
modeled using the Incremental Model in order to obtain a linear
optimization problem.
cp (15)
4.3. Optimization results
(16)
The degradation suffered by the battery of an EV (Electric
Vehicle) during a full year is compared using the proposed algo-
rithm and other charging strategies. To this purpose, the following
scenario is evaluated. The average daily temperature for the entire
year of 2011 (8b) was obtained for Madrid, Spain from Ref. [31]. The
EV profile used to simulate the EV usage can be seen in Fig. 8a.
Three different charging methods were simulated for comparison
(17) purposes. The first method is an “on-demand” charge, in other
words, there is no energy management of the electric vehicle. The
battery is charged as soon as it is plugged into the grid until it is
fully charged. The second mode is a simple optimization problem,
where the objective function (7) is changed to minimize the sum of
the depth-of-discharges suffered throughout the day. This method
cp
is similar to the one proposed by Riffonneau et al. [19], where the
(18) degradation is calculated proportional to the depth-of-discharge.
42 I.J. Fernández et al. / Energy 60 (2013) 35e43

Daily aggregated consumption of EV driver Average temperature for Madrid during 2011
8000 40

7000 35

6000
30

Temperature [ C]
5000
SOC [KWh]

25
4000
20
3000
15
2000

1000 10

0 5
0 5 10 15 20 25 0 50 100 150 200 250 300 350 400
time [h] time [day]
(a) User profile used for simulation (b) Average temperature profile used for simulation
Fig. 8. Profiles used for simulation purposes.

The third method tested for energy management was the proposed
algorithm, minimizing the degradation suffered by the battery
throughout the day. The “on-demand” charging method has been
used as a reference point when presenting the results for the im-
provements achieved by the two models. The results for the
different energy management strategies can be seen in Table 2;
results show a 48% improvement when minimizing battery
degradation, instead of only minimizing the depth-of-discharges,
saving the electric vehicle owner about 958V annually (for a bat-
tery cost of 15000V).
Both strategies for optimal energy management, at a given
constant temperature, have the same Ah-throughput throughout
the day. That is, both algorithms penalize high depth-of-discharges
throughout the day equally, obtaining similar battery degradations.
The difference between both charging methods is emphasized
when there is a change in average temperature between two
consecutive periods. The proposed algorithm will tend to optimally
shift the charges to cause the least degradation possible according
to the battery degradation model proposed in Section 3. The energy
management algorithm has been validated using the daily average
temperature throughout the entire year of 2011, but the algorithm
could be validated discriminating between the average tempera-
ture during the day and during the night, which would lead to
better results. The number of piecewise linear functions does not
affect the performance of the optimization model; on the contrary,
the maximum number of segments that the piecewise linear
functions have will affect the outcomes. The optimization model
could include smaller intervals between temperatures, obtaining
more accurate results.
Table 2
Results: average degradation improvement and money saved in one year, using an
‘On-Demand’ scenario as reference.
5. Conclusion
Using the data obtained experimentally from several Li-ion
batteries, the average temperature and the depth-of-discharge
have been identified as the variables that affect battery degrada-
tion the most. This was done using self-organizing maps, obtaining
results which clearly identify both variables as the two main con-
tributors to battery degradation. A battery degradation model was
later computed using these two variables as inputs.
An intelligent energy management system is of great interest for
the owners of electric vehicles. Battery degradation models allow
the optimal management of the charge, and even the discharge of
the electric vehicles, increasing the overall expected battery life. A
model allows researchers to evaluate different strategies, and to
study the feasibility of different projects concerning energy man-
agement of electric batteries. A possible optimal energy manage-
ment system was formulated and evaluated, comparing the
degradation suffered by the battery in one year using three
different charging strategies. Results show that considering the
temperature during the optimization problem yields considerably
better results than just minimizing the sum of the depth-of-
discharge throughout the day.
Future work involves the validation of the current degradation
model, comparing results with other lithium-ion battery cells.
Different capacities besides of different manufacturers should be
considered for experimentation. On the other hand, using lithium-
ion battery cells with different cathodes may be of interest as well,
in order to analyze the effect that this may have over the overall
degradation as a function of temperature and depth-of-discharge.
The battery degradation model proposed in this paper is specific
to the lithium-ion battery PL-9059156, but future work may
extrapolate the model to different battery chemistries.

Minimizing Minimizing battery References

depth-of-discharge degradation
[1] Smith WJ. Stochastic optimal charging of electric-drive vehicles with renew-
Battery degradation 4.32% 6.39% able energy. Energy 2010;35(12):4514e21.
Money saved 647V 958V [2] Pantos M. Stochastic optimal charging of electric-drive vehicles with renew-
able energy. Energy 2011;36(11):6567e76.
[3] Kiviluoma J, Meibom P. Methodology for modelling plug-in electric vehicles in
the power system and cost estimates for a system with either smart or dumb
Table 3 electric vehicles. Energy 2011;36(3):1758e67.
Absolute mean degradation error for different temperatures and DOD (Ah). [4] Shafie-khah M, Moghaddam MP, Sheikh-El-Eslami M Kazem, Rahmani-
Andebili M. Modeling of interactions between market regulations and
T z 25  C T z 5 C T z 50  C
behavior of plug-in electric vehicle aggregators in a virtual power market
DOD z 10% 51.5589% 73.4802% 49.1146% environment. Energy 2012;40(1):139e50.
DOD z 45% 63.7911% 64.2588% 44.3272% [5] Sun F, Hu X, Zou Y, Li S. Adaptive unscented kalman filtering for state of
DOD z 80% 58.3881% 51.5718% 61.7571% charge estimation of a lithium-ion battery for electric vehicles. Energy
2011;36(5):3531e40.
 I.J. Fernández et al. / Energy 60 (2013) 35e43
[6] Marano V, Onori S, Guezennec Y, Rizzoni G, Madella N. Lithium-ion batteries
life estimation for plug-in hybrid electric vehicles. In: IEEE vehicle power and
propulsion conference 2009. p. 536e43.
[7] Somogye R. Aging model of ni-mh batteries for use in hybrid-electric vehicles.
Ph.D. dissertation. The Ohio State University; 2004.
[8] Ning G, Popov BN. Cycle life modeling of lithium-ion batteries. Journal of the
Electrochemical Society 2004;151:A1584.
[9] Li J, Murphy E, Winnick J, Kohl PA. Studies on the cycle life of commercial
lithium ion batteries during rapid charge-discharge cycling. Journal of Power
Sources 2001;102(1e2):294e301.
[10] Santhanagopalan S, Guo Q, Ramadass P, White RE. Review of models for
predicting the cycling performance of lithium ion batteries. Journal of Power
Sources 2005;156(2):620e8.
[11] Johnson VH. Battery performance models in ADVISOR. Journal of Power
Sources 2002;110(2):321e9.
[12] Zi-lin M, Xao-Jian M, Jun-xi W, Jia-xi Q, Bin Z. Research on SOC estimated
strategy of Ni/MH battery used for hybrid electric vehicle. In: Vehicle power
and propulsion conference, 2008. VPPC’08. IEEE. IEEE; 2008. p. 1e4.
[13] Chen M, Rincon-Mora G. Accurate electrical battery model capable of pre-
dicting runtime and IV performance. Energy Conversion, Ieee Transactions on
2006;21(2):504e11.
[14] Shen WX, Chau KT, Chan CC, Lo EWC. Neural network-based residual capacity
indicator for nickel-metal hydride batteries in electric vehicles. IEEE Trans-
actions on Vehicular Technology 2005;54(5):1705e12.
[15] Magazine S, Policy P, Grid S. Modeling lithium ion battery degradation in
electric vehicles. In: 2010 IEEE conference on innovative technologies for an
efficient and reliable electricity supply (CITRES) 2010. p. 349e56.
[16] Drouilhet S, Johnson B. A battery life prediction method for hybrid power
applications. Contract DE-AC36-83CH10093. Chicago, IL: US Dept. Energy;
1997.
[17] Juul N. Battery prices and capacity sensitivity: electric drive vehicles. Energy
2012;47(1):403e10.
[18] Li Z, Ouyang M. The pricing of charging for electric vehicles in china-dilemma
and solution. Energy 2011;36(9):5765e78.
43
[19] Riffonneau Y, Bacha S, Barruel F, Ploix S. Optimal power flow management for
grid connected pv systems with batteries. 1e1. IEEE Transactions on Sus-
tainable Energy 2011;(99).
[20] Sioshansi R, Denholm P. The value of plug-in hybrid electric vehicles as grid
resources. The Energy Journal 2010;31(3):1e23.
[21] Sandels C, Franke U, Ingvar N, Nordstrom L, Hamren R. Vehicle to grid: Monte
Carlo simulations for optimal aggregator strategies. In: Int. conf. on power
system technology. IEEE; 2010. p. 1e8.
[22] Saber AY, Venayagamoorthy GK. Intelligent unit commitment with vehicle-to-
grideA cost-emission optimization. Journal of Power Sources 2010;195(3):
898e911.
[23] Saber A, Venayagamoorthy G. Unit commitment with vehicle-to-grid using
particle swarm optimization. In: PowerTech, 2009 IEEE Bucharest. IEEE; 2009.
p. 1e8.
[24] Guille C, Gross G. A conceptual framework for the vehicle-to-grid (V2G)
implementation. Energy Policy 2009;37(11):4379e90.
[25] Premanand R, Haran B, White R, Popov B. Capacity fade studies on spinel
based Li-ion cells. In: Battery conference on applications and advances, 2002.
The seventeenth annual. IEEE; 2002. p. 25e30.
[26] Ning G, Haran B, Popov BN. Capacity fade study of lithium-ion batteries cycled
at high discharge rates. Journal of Power Sources 2003;117(1e2):160e9.
[27] Haifeng D, Xuezhe W, Zechang S. A new SOH prediction concept for the power
lithium-ion battery used on HEVs. In: Proc. of IEEE vehicle power and pro-
pulsion conference VPPC 2009. p. 1649e53.
[28] Danilov D, Notten P. Adaptive battery management systems for the new
generation of electrical vehicles. In: Vehicle power and propulsion conference,
2009. VPPC’09. IEEE. IEEE; 2009. p. 317e20.
[29] Serrao L, Chehab Z, Guezennee Y, Rizzoni G. An aging model of Ni-MH bat-
teries for hybrid electric vehicles. In: Vehicle power and propulsion, 2005 IEEE
conference. IEEE; 2005. p. 78e85.
[30] Belt J. Battery test manual for plug-in hybrid electric vehicles. Tech. Rep. Idaho
National Laboratory (INL); 2008.
[31] Tutiempo network, s.l. Visited May 2012. [Online]. Available: http://www.
tutiempo.net/.