Applied Energy 113 (2014) 189–198

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

A dynamic model-based estimate of the value of a vanadium redox flow

battery for frequency regulation in Texas
Robert L. Fares ⇑, Jeremy P. Meyers 1, Michael E. Webber
Department of Mechanical Engineering, The University of Texas, Austin, TX 78712, United States

h i g h l i g h t s

 A model is implemented to describe the dynamic voltage of a vanadium flow battery.

 The model is used with optimization to maximize the utility of the battery.
 A vanadium flow battery’s value for regulation service is approximately $1500/kW.

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

Article history: Building on past work seeking to value emerging energy storage technologies in grid-based applications,
Received 18 February 2013 this paper introduces a dynamic model-based framework to value a vanadium redox flow battery (VRFB)
Received in revised form 14 May 2013 participating in Texas’ organized electricity market. Our model describes the dynamic behavior of a VRFB
Accepted 12 July 2013
system’s voltage and state of charge based on the instantaneous charging or discharging power required
Available online 7 August 2013
from the battery. We formulate an optimization problem that incorporates the model to show the poten-
tial value of a VRFB used for frequency regulation service in Texas. The optimization is implemented in
Keywords:
Matlab using the large-scale, interior-point, nonlinear optimization algorithm, with the objective func-
Energy storage
Flow battery
tion gradient, nonlinear constraint gradients, and Hessian matrix specified analytically. Utilizing market
Economics prices and other relevant data from the Electric Reliability Council of Texas (ERCOT), we find that a VRFB
Frequency regulation system used for frequency regulation service could be worth approximately $1500/kW.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction In recent years, high-performance electrochemical energy

In 2010, U.S. electric providers sold over 3.7 trillion kW h of elec-
trical energy, generating nearly $369 billion in revenue [1]. Despite
the enormity of the electricity industry, the U.S. electric grid has very
little capacity to store electricity. In 2011, the U.S. grid had only
22 GW of electric energy storage capacity, compared with over
1000 GW of generation capacity [1,2]. Because of the high cost of
conventional electricity storage technologies (such as pumped-hy-
dro energy storage [3]), it is typically more economical to generate
electricity ‘‘on demand,’’ generating and then delivering it to the
end user in real time. To reliably deliver electricity on demand, gen-
eration, transmission, and distribution equipment are sized to serve
peak electric load. Furthermore, electric generators set aside capac-
ity for grid ancillary services to reliably balance electric supply and
demand in real time. These two aspects of the present electric grid
represent a non-trivial component of the cost of electricity.
⇑ Corresponding author.
E-mail address: robertfares@mail.utexas.edu (R.L. Fares).
1
Present address: EnerVault Corporation, 1244 Reamwood Avenue, Sunnyvale, CA
94089, United States.
storage technologies such as sodium–sulfur, lithium–ion, and redox
flow batteries (RFBs) have been developed to support grid applica-
tions. Despite advances in electrochemical energy storage technol-
ogy, batteries have only been sparingly implemented on the U.S.
electric grid. In 2011, there were less than 140 MW of batteries
installed [2]. This fact can be attributed to the high cost of existing
battery systems and uncertainty of their long-term reliability. Fur-
thermore, there is uncertainty about the precise economic value of
battery energy storage in grid-level applications. There are primar-
ily two reasons for this uncertainty. First, unlike a traditional com-
modity storage facility, a battery cannot store electricity without
losing some energy to conversion losses and other inefficiencies.
Second, the value of energy storage in a grid application is directly
affected by the external price of energy. Whether batteries are sell-
ing wholesale energy during peak hours or working to balance and
control the electric grid, the energy they use to perform these
services must come from somewhere, and the cost of this energy
has a marked effect on the value of battery energy storage as it
participates in the electricity marketplace. For these reasons, an
assessment of the potential value of grid-based battery energy
storage would benefit from two features: (1) an appropriate battery

0306-2619/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.apenergy.2013.07.025
190 R.L. Fares et al. / Applied Energy 113 (2014) 189–198
energy storage model to describe the particular capabilities of the
battery in question and (2) a framework for showing how the bat-
tery could operate to produce the most value in an application.
The goal of this paper is to assess the potential value of a vana-
dium redox flow battery (VRFB) for frequency regulation service in
Texas. Many researchers have highlighted the technical compati-
bility between electrical energy storage and frequency regulation
service [4–7]. In order to show the value of a VRFB for regulation
service, we perform a time-domain analysis of a VRFB’s participa-
tion in Texas’ organized electricity market that includes the detail
of time-varying regulation capacity prices. We describe the instan-
taneous energy conversion capabilities of a VRFB using a dynamic,
control-oriented battery model. To show how a VRFB could operate
economically for regulation service, we incorporate this model into
an economic optimization routine. The following section discusses
past work that has assessed the value of energy storage in grid-
based applications. Section 3 discusses the development of a con-
trol-oriented VRFB model, and how we implement this model with
optimization. Section 4 uses the results of the optimization to as-
sess the value of a VRFB for regulation service in Texas. Finally, Sec-
tion 5 summarizes our work and discusses possible future
research.
2. Background
Because electric energy storage has the potential to reduce util-
ity costs associated with peak demand and power system control, a
number of researchers have sought to demonstrate the benefits of
grid-based battery energy storage. Early investigations focus on
lead-acid battery applications for electric utilities [8,9]. These stud-
ies use utility-level knowledge to assess the economic benefit of
grid-based energy storage. In recent years, organized competitive
electricity markets have emerged in place of conventional verti-
cally-integrated electric utilities. Rather than a single entity con-
trolling a portfolio of generation, transmission, and distribution
resources, competitive electricity markets permit diverse parties
to offer their electric generation resources into a wholesale power
marketplace. Numerous studies have highlighted the unique
opportunity that these new markets present for energy storage
[10–13]. Early studies by researchers from the U.S. Department
of Energy’s Energy Storage Systems Program identify new opportu-
nities for storage in competitive electricity markets [10,11]. Fol-
lowing these studies, there have been detailed investigations of
energy storage operating in the New York Independent System
Operator (NYISO) market [12] and the California Independent Sys-
tem Operator (CAISO) market [13]. Both of these studies identify
frequency regulation as a high-value application for grid-based en-
ergy storage [12,13]. A separate study of energy storage operating
in the NYISO market analyzes the economics of storage for whole-
sale electricity arbitrage and regulation service [7]. It shows that
energy storage for frequency regulation is more valuable than stor-
age for energy arbitrage—even in transmission-constrained New
York City [7]. Furthermore, studies of battery energy storage oper-
ating in the German electricity market identify frequency regula-
tion as the highest-value application for battery energy storage
[14,15]. We build on these studies by performing a time-domain
analysis of the potential value of a VRFB used for regulation service
from 2007–2009 in Texas’ organized electricity market, which is
administered by the Electric Reliability Council of Texas (ERCOT).
By doing so, we capture the daily character of chronological regu-
lation capacity prices and wholesale energy prices, and demon-
strate how a VRFB could respond to those price signals.
Many estimates of the value of energy storage use a
model-based time-domain analysis to show how a battery could
participate in the electricity market [7,16–18]. The energy storage
capabilities of a battery are conventionally described using a sim-
ple, black-box energy storage model, which describes a battery as a
container for energy with regular losses anytime energy is added to
or removed from the battery. With this sort of model, time-domain
analysis of electricity market power flows is conducted by tracking
the amount of energy stored in the device as it performs its duty on
the grid. At each time step, k, the instantaneous amount of energy
stored in the battery is tracked as a function of the power flowing
to or from the battery as shown in the following equations [18]:
pffiffiffiffiffiffiffiffiffi
Ebatt;out ðkÞ ¼ P batt;out ðkÞDt= gbatt ð1Þ
pffiffiffiffiffiffiffiffiffi
Ebatt;in ðkÞ ¼ P batt;in ðkÞDt gbatt ð2Þ
Ebatt ðkÞ ¼ Ebatt ðk  1Þ þ Ebatt;in ðkÞ  Ebatt;out ðkÞ ð3Þ
The variable Pbatt,out(k) is the flow of power out of the battery;
Pbatt,in(k) is the flow of power into the battery; and Ebatt(k) is the
quantity of energy stored in the battery at each time step of dura-
tion Dt. The symbol gbatt represents the round-trip efficiency of
the battery, and is treated as constant. The square root of gbatt is
used in Eqs. (1) and (2) so that energy losses are imposed equally
on discharging and charging power, respectively.
This model has been used for operational management of a bat-
tery using optimization [16,17]. The objective of the optimization
problem is a function of the power flowing in and out of the battery
at each time step (Pbatt,in(k), Pbatt,out(k)) and a relevant electricity-
market price signal. Bounds (see Eqs. (4)–(6)) are placed on the
variables to find an optimal operation strategy for the battery that
does not violate its technical limits.
8k P batt;out ðkÞ 6 Pmax;out ð4Þ
8k P batt;in ðkÞ 6 P max;in ð5Þ
8k 0 6 Ebatt ðkÞ 6 Ecapacity ð6Þ
The conventional model of a battery interfacing with the elec-
tric grid is limited because it tracks the real-time state of a battery
with only one variable: Ebatt(k). This energy state variable effec-
tively assumes a battery operates at a constant voltage. Because
voltage varies with hysteresis and nonlinearly with a battery’s duty
cycle [19], the conventional model described in Eqs. (1)–(3) is an
abstraction of a battery’s real-time behavior. It is more accurate
to consider voltage variations and decompose a battery’s stored
energy into two state variables: the terminal voltage and the state
of charge (SOC), defined in Eq. (7) as the ratio of stored charge (q)
to charge capacity (qmax).
SOC ¼ q=qmax ð7Þ
While the conventional model is useful for energy storage eco-
nomic analysis, its black-box description of a battery’s state leads
to difficulties in connecting operational strategies gleaned from
the conventional model with real-time battery control. To over-
come the shortcomings of the conventional model, we develop a
control-oriented battery energy storage model, which encapsulates
the real-time energy conversion capabilities of a battery within a
dynamic relationship between current, voltage, and SOC. With this
model, we describe battery voltage variations within an economic
optimization framework, and help to connect system-level eco-
nomic analysis with real-time optimal control of grid-connected
battery energy storage.
 R.L. Fares et al. / Applied Energy 113 (2014) 189–198
3. Methods
3.1. Development of a control-oriented dynamic battery model
The electrochemical reactions inside of a battery are bound by
conservation of species during repeated cycling and, neglecting
parasitic losses, a battery conserves electric charge. Thus, the
amount of charge stored inside a battery can be approximated with
reasonable accuracy by coulomb counting: integrating the flow of
current in and out of a battery as it operates. We track the state
of charge of a battery (defined in Eq. (7)) using this method.
The voltage of a battery is a measure of the potential difference
between its positive side (cathode) and negative side (anode). In
order for a battery to take in electric current, an external power
source is applied to increase its voltage. Similarly, in order for a
battery to produce current, an external load is applied to decrease
the voltage. The amount of applied voltage required to produce a
specified electrical current depends on the instantaneous internal
composition of the battery (SOC) and the desired rate of charge
or discharge [19].
By modeling a battery using both its voltage and SOC, we de-
scribe a battery’s voltage-dependent energy efficiency within a dy-
namic relationship between voltage, SOC, and charging or
discharging current. Our model is based on Manwell and McGo-
wan’s dynamic battery energy storage model, which was devel-
oped for quasi-steady, time-series simulations of power-
producing systems that incorporate battery energy storage, such
as wind-diesel hybrid energy systems [20]. Manwell and McGo-
wan’s model describes a battery as a voltage source in series with
an internal resistance, as shown in Fig. 1.
The terminal voltage of the battery (V) is a function of the inter-
nal voltage (E), the input current (I), and the internal resistance
(R0), as given in Eq. (8). The internal potential source (E) varies with
SOC, and the internal resistance (R0) is modeled as constant. Dis-
charging current is given a positive sign.
V ¼ E  IR0 ð8Þ
Their model assumes a galvanostat is placed between the battery
and the load, so that the model takes the instantaneous current load
on the battery as a stepwise-constant input.
We adapt Manwell and McGowan’s battery model to describe
the capabilities of a VRFB within an economic battery-manage-
ment optimization problem. Using Eq. (8), we describe how a
VRFB’s terminal voltage varies as its SOC changes, and as the ap-
plied current changes. In doing so, we describe the state-depen-
dent ratio between stored charge (SOC) and available electric
energy. We assume that all of the electric charge stored in external
electrolyte tanks is available for reaction. While mass-transport
losses will ultimately limit performance at very high rates of
charge or discharge, or as the concentration in the tanks is depleted
at the end of a charging or discharging cycle, we limit operation to
rates well below the mass transport limit, and constrain the SOC to
a range where the concentrations do not begin to vary by orders of
Fig. 1. Manwell and McGowan’s model describes a battery as a variable potential
source in series with a constant internal resistance. Adapted from [20].
191
magnitude as material is further depleted. The limits of applicabil-
ity of the model are discussed further below. We describe the
state-dependent potential function of a VRFB modeled as in Fig. 1
using the following equation:
EðtÞ ¼ E0 þ AðSOCðtÞÞ ð9Þ
E0 and A in Eq. (9) are parameters whose values can be extracted
from galvanostatic charging and discharging data. While this linear
voltage model does not capture nonlinear voltage behavior near 0%
and 100% SOC, its linearity is advantageous for the purposes of
optimization. Unlike Manwell and McGowan’s model, we describe
the potential source using the same function (Eq. (9)) for charging
and discharging. This assumption is consistent with the overpoten-
tial behavior of a VRFB, and it minimizes the complexity of the
model for the purposes of integration with optimization.
In order to estimate the value of the parameters in the VRFB
model (E0, A, and R0) we utilize galvanostatic charging/discharging
data from a kW-scale integrated VRFB system. Zhao et al. per-
formed a number of experiments to characterize a 1 kW stack of
14 vanadium redox flow cells connected electrically in series
[21]. Fig. 2 shows the voltage behavior of the VRFB stack during
a complete charge/discharge cycle at three rates. Using the VRFB
voltage data in Fig. 2, the internal resistance parameter, R0, in Eq.
(8) can be estimated by examining the rate of change of the stack
voltage with current at a constant state of charge as shown in the
following equation:
@V
 ¼ R0 ¼ 15 mX ð10Þ
@I
Fig. 3 shows how VRFB stack voltage varies with rate of charge/
discharge at 50% SOC.
With the value of R0 determined, the value of the parameters E0
and A in Eq. (9) is estimated via a least-squares curve fit of Eq. (8)
to experimental data. We fit Eq. (8) to experimental discharging
voltage data at 60 mA/cm2 taken where the VRFB voltage is
approximately linear, between 20% and 80% SOC. Based on the
curve fit in Fig. 4, we estimate the parameters E0 and A for the
1 kW VRFB stack as given in the following equations:
E0 ¼ 18:5 V ð11Þ
A ¼ 2:69 V ð12Þ
In order to demonstrate the validity of the model developed in
this section, we compare experimental data collected by Zhao et al.
[21] to galvanostatic curves predicted by the voltage model of Eq.
(8). Fig. 5 compares the voltage model to discharging test data at
50 mA/cm2 and Fig. 6 compares the voltage model to charging test
data at 60 mA/cm2. As shown in Figs. 5 and 6, the linear voltage
model developed in this section describes both the charging and
24
50 mA/cm2
22 60 mA/cm2
70 mA/cm2
Stack Voltage (V)
20
18
16
14
12
0 10 20 30 40
Test Time (min)
Fig. 2. Zhao et al. collected galvanostatic voltage data from a 1 kW VRFB stack
consisting of 14 flow cells connected electrically in series. Adapted from [21].
192 R.L. Fares et al. / Applied Energy 113 (2014) 189–198

24
Experimental Data
20.5 Voltage Model
22
Stack Voltage (V)

Stack Voltage (V)

20
20.0 R0 ≈ 15 mΩ
18

19.5 16

14
19.0
12

−60 −40 −20 0 20 40 60 0.0 0.2 0.4 0.6 0.8 1.0

Stack Current (A)
State of Charge
Fig. 3. The voltage behavior of a VRFB with varying current is approximately linear
Fig. 6. The voltage model of Eqs. (8) and (9) is validated by comparison to
at a constant SOC. Adapted from [21].
experimental charging data at 60 mA/cm2 collected from a 1 kW VRFB cell stack in
[21].

20.0
Experimental Data
Curve Fit

19.5
Stack Voltage (V)

19.0

18.5

18.0
0.3 0.4 0.5 0.6 0.7
Fig. 7. The VRFB voltage model is used to estimate SOC with battery power as an
State of Charge
input. The voltage of the battery is tracked internally for the purposes of the
estimate.
Fig. 4. The flow-battery voltage model of Eqs. (8) and (9) is fit to experimental
discharging data at 60 mA/cm2 collected from a 1 kW VRFB cell stack in [21].

The model devised up to this point describes the energy losses

of a VRFB cell stack brought about by internal, electrochemical
Experimental Data
Voltage Model
phenomena. In addition to energy losses within the battery itself,
20
a grid-connected VRFB is subject to losses from necessary power
conditioning equipment, and from power required to operate
Stack Voltage (V)

18
16
14
12
0.0 0.2 0.4 0.6
State of Charge
Fig. 5. The voltage model of Eqs. (8) and (9) is validated by comparison to
experimental discharging data at 50 mA/cm2 collected from a 1 kW VRFB cell stack
in [21].
discharging voltage of a VRFB at different rates of charge/discharge
and at different levels of SOC. By describing the voltage behavior of
a VRFB, the model captures the state-dependent energy-conver-
sion capabilities of a VRFB under diverse operating conditions.
Fig. 7 shows how the battery model developed in this section is
used to estimate SOC based only on the flow of power to and from
a VRFB. At each time step, the input battery power is divided by the
battery voltage to find the current flowing through the battery. The
flow of current is summed up to estimate the battery’s SOC. Finally,
the battery current and SOC are plugged into the voltage model of
Eqs. (8) and (9) to estimate the voltage of the battery at the begin-
ning of the next time step. We use this model within an optimiza-
tion routine to estimate SOC based on market power flows, so that
an economic operational strategy can be developed that does not
exceed the specified limits of the battery in question.
pumps and other auxiliary plant equipment. Because a VRFB oper-
ating in the electricity market interfaces with the electric grid, its
DC power must be converted to AC before it is injected into the grid
and vice versa. We assume a one-way power inverter/rectifier effi-
ciency of 93% based on recommendations in the literature [22,23].
To approximate the level of auxiliary power required, we utilize
experimental VRFB data collected at Risø DTU National Laboratory
0.8 1.0 for Sustainable Energy [23]. This data show that a 15 kW VRFB sys-
tem requires approximately 1.4–1.6 kW of auxiliary power while it
is running, so we assume a constant auxiliary power requirement
for a VRFB equal to 10% of the system’s rated power. Eq. (13) de-
scribes the relationship between the power seen by the grid (Pgrid)
and the power seen by the VRFB (Pbatt). We give discharging power
a positive sign.
8
< ginv erter Pbatt  0:10Prated
> 1
if charging
Pgrid ¼ ginv erter Pbatt  0:10Prated if discharging ð13Þ
>
:
Pbatt ¼ 0 if idle
The VRFB model developed in this section has sufficient com-
plexity to describe the instantaneous energy-conversion capabili-
ties of a VRFB within an economic optimization routine.
However, it has a number of limitations one should consider before
implementing the model. Because it assumes a linear voltage pro-
file to reduce computational complexity, the model does not de-
scribe the asymptotic behavior of a battery’s voltage near 0% and
100% SOC. Furthermore, the model assumes that no charge is lost
to parasitic electrochemical processes or electrolyte crossover dur-
ing cycling. Moreover, the model does not describe the long-term
 R.L. Fares et al. / Applied Energy 113 (2014) 189–198 193

consequences of repeated cycling that might affect VRFB perfor-
mance. This paper does not directly address optimal pump opera-
tion [24], electrolyte thermal management, and power electronics
improvements that could affect the performance of the energy
storage system. Nevertheless, the model has a number of advanta-
ges over typical storage models traditionally utilized for time-do-
main economic analysis of grid-based energy storage. In the
following section, we develop an optimization problem for eco-
nomic operational management of a VRFB for frequency regulation
service in ERCOT, placing appropriate constraints on the VRFB to
minimize the effect of the model’s limitations.
3.2. Economic management of a VRFB for frequency regulation in
Texas
To develop an effective optimization problem to glean the po-
tential value of a VRFB used for frequency regulation in ERCOT, it
is important to first discuss how ERCOT organizes a market for reg-
ulation service, and controls the frequency of the electric grid in
real time. ERCOT operates an organized wholesale energy market
to distribute energy from independent generators to electricity
customers in a technically and economically efficient way. In dis-
tributing this energy, ERCOT takes on the role of the grid operator,
and is tasked with ensuring that each electricity customer has a
reliable source of electricity at his/her particular connection to
the electric grid. To do so, ERCOT organizes a market for ancillary
services, including frequency regulation service. Electric generators
offer some of their power capacity into the market for regulation
service each hour. ERCOT divides regulation service into up regula-
tion service (URS) (committing to increase power output on de-
mand), and down regulation service (DRS) (committing to
decrease power output on demand). Based on appropriate fore-
casts, ERCOT procures a specified amount of the power offered
for URS and DRS each hour of the day, and sets a clearing price in
$/MW for up and down regulation capacity. After procuring electric
(NERC) control performance standards (CPS1, CPS2) [28] to gener-
ate these signals and automatically dispatch generation in re-
sponse to frequency deviations. These historic signals allow us to
gather how a battery might be dispatched for frequency regulation
in the future. We assume that a VRFB is dispatched at the level of
power it has committed for regulation service, and in the direction
of the aggregate ERCOT automatic generator control signal. The
binary parameter, y, represents the direction of regulation require-
ments, and is discretized into three-minute time steps with index
k. Eq. (14) defines y(k).

1 if URS required
yðkÞ ¼ ð14Þ
0 if DRS required
The optimization routine operates a day ahead. The decision
variables in the optimization problem are the amount of battery
power offered to the grid operator for up-regulation service (URS)
and down-regulation service (DRS). In ERCOT, these offers are
made hourly [29], so the optimization considers 48 decision vari-
ables (24 hourly URS offers and 24 hourly DRS offers). The VRFB
state variables depend directly on the value of these decision vari-
ables and the parameter y(k). They are tracked in three-minute
time steps within the optimization problem. Table 1 summarizes
the variables considered in each day-ahead optimization problem.
We relate the 1920 dependent variables to the 48 decision vari-
ables using the equality constraints of Eqs. (15)–(18), which inter-
nalize the model schematic of Fig. 7 within the economic
optimization routine.
Pbatt ðkÞ ¼ URSðhÞyðkÞ þ DRSðhÞð1  yðkÞÞ ð15Þ
IðkÞ ¼ P batt ðkÞ=Vðk  1Þ ð16Þ
1 X
k
SOCðkÞ ¼ SOC i  IðiÞDt ð17Þ
qmax
i¼1

generators for regulation service, ERCOT dispatches those genera-

tors as necessary to control the frequency of the electric grid, rap- VðkÞ ¼ E0 þ A SOCðkÞ  IðkÞR0 ð18Þ
idly adjusting the level of power supply to match demand. During
To ensure the VRFB operates within its limits while it provides fre-
dispatch, energy delivered to the grid is credited at the prevailing
quency regulation service, we bound the variables as shown in
wholesale price of energy. A VRFB procured for regulation service
Table 2.
would be credited at the prevailing clearing price of regulation
We bound the amount of power the VRFB can offer into the
capacity, and then ordered to alternately charge and discharge to
marketplace for regulation service within the rated power of the
balance grid frequency. Importantly, the VRFB would be debited
hypothetical VRFB system we want to test, and we bound the VRFB
for charging energy and credited for discharging energy at the pre-
SOC within 20% and 80%. In this range, our model describes the lin-
vailing wholesale energy price during dispatch.
ear voltage of the VRFB well. Furthermore, in this range the VRFB
To estimate the potential value of a VRFB for frequency regula-
voltage is not near its minimum or maximum permissible value,
tion in Texas, we incorporate our control-oriented VRFB model
and it is unlikely that extraneous charge-transfer reactions will oc-
with optimization to show how a VRFB might operate economi-
cur. This constraint also simplifies the control system for the bat-
cally for frequency regulation in ERCOT. For the purposes of our
tery and therefore represents a likely operation mode for a real-
estimate, we assume perfect day-ahead foresight of relevant en-
life RFB. By bounding the SOC within this range, we improve the
ergy prices and regulation capacity prices, which ERCOT publishes
accuracy of our model. The bounds on the VRFB voltage and cur-
publicly [25]. Furthermore, we assume perfect foresight of the ER-
rent are inherited from the bounds on power and SOC. In addition
COT frequency-regulation dispatch signal. Although these assump-
tions cause our model to overestimate the potential value of a
VRFB, a study by Townsend shows that existing methods of price Table 1
forecasting allow storage to obtain approximately 95% of perfect- 1968 Variables are considered in each day-ahead optimization problem for frequency
knowledge operating profits [26]. regulation.

Presently, how battery energy storage might be automatically Variable Symbol Time resolution Variables per day
dispatched for frequency regulation is not perfectly understood. Decision variables
Because a battery can respond to ERCOT control signals more rap- URS offer (MW) URS 1h 24
idly than any traditional thermal generator, utilization of battery DRS offer (MW) DRS 1h 24
energy storage could lead to less procurement of regulation re- Dependent variables
sources, and tighter control of grid frequency [5,6,27]. For the pur- Battery power (MW) Pbatt 3 min 480
poses of our analysis, we utilize historic automatic generator Battery voltage (V) V 3 min 480
Battery current (A) I 3 min 480
control signal data from 2007–2009 provided to us by ERCOT. ER-
Battery SOC SOC 3 min 480
COT follows North American Electric Reliability Corporation
194 R.L. Fares et al. / Applied Energy 113 (2014) 189–198

Table 2 X
24
The variables considered in the optimization problem are bounded so that the VRFB Obj ¼ ðURSðhÞ  0:129Prated ÞC up ðhÞ  ðDRSðhÞ
considered operates within its limits. h¼1

Variable Lower bound Upper bound X

480
þ 0:129Prated ÞC down ðhÞ þ ðPbatt ðkÞ
URS 0 Prated k¼1
DRS Prated 0
Pbatt Inherited Inherited  0:129Prated ÞDtkðkÞ ð22Þ
V Inherited Inherited
I Inherited Inherited We implement the optimization in Matlab using the large-scale,
SOC 20% 80% interior-point, nonlinear optimization algorithm. We specify the
objective function gradient, nonlinear constraint gradients, and
Hessian matrix analytically. Using locational wholesale energy
to the bounds of Table 2, we also bound the end-of-day SOC as in prices from ERCOT’s North, South, West, and Houston hubs, we
Eq. (19). This constraint ensures that the VRFB’s initial energy is re- solve the optimization problem for each day of 2007–2009. We
placed by the end of each day, so that the potential for future rev- consider a nominal 1 MW VRFB system consisting of 100 VRFB
enue is protected. modules connected electrically in parallel. Each of these modules
consists of 140 VRFB cells connected electrically in series. Each of
SOCðk ¼ 480Þ P 50% ð19Þ the cells has an electrode geometric surface area of 875 cm2, con-
sistent with the VRFB stack characterized by Zhao et al. [21]. We
The objective function of the optimization problem is the total
assume that the VRFB cells characterized by Zhao, et al. can be
daily revenue from providing frequency regulation service in ER-
lumped together to model a MW-scale VRFB system. To show
COT, and it is maximized by the optimization routine. Eq. (20) de-
how the energy capacity of a VRFB affects its value, we consider
fines the objective function.
systems ranging in discharge duration from 30 min to 4 h. The dis-
X
24 charge duration of a battery indicates how long it can discharge at
Obj ¼ URSgrid ðhÞC up ðhÞ  DRSgrid ðhÞC down ðhÞ its rated power before it is fully depleted.
h¼1

X
480
þ Pgrid ðkÞDtkðkÞ
k¼1
In Eq. (20), Cup(h) and Cdown(h) are the hourly up-regulation and
down-regulation prices in ERCOT, respectively. The parameter k(k)
is the prevailing price of wholesale energy in ERCOT at time step
k. The variables URSgrid(h) and DRSgrid(h) are the up and down regu-
lation offers seen by the grid each hour of the day, and Pgrid(k) is the
VRFB system power seen by the grid at each time step k. As outlined
in Eq. (13), the power seen by the grid differs from the power seen
by the VRFB. We incorporate this difference in the objective func-
tion. Because Eq. (13) is conditional and highly non-smooth, we
approximate Eq. (13) using Eq. (21). By removing an objective func-
tion discontinuity, this assumption greatly simplifies the optimiza-
tion problem without marked error. Fig. 8 compares Eqs. (13) and
(21).
Pgrid  Pbatt  0:129Prated
It should be noted that the estimate of Eq. (21) is only used within
the optimization procedure. In Section 4, Eq. (13) is used to adjust
the optimization results for energy losses to estimate VRFB revenue.
Eq. (22) is the objective function of the optimization problem with
Eq. (21) incorporated.
1.0
Equation 13
Equation 21
0.5
Pgrid Prated
ð20Þ
4. Results and discussion
The optimization problem formulated in the previous section
requires approximately 22 s of computation time to determine
an optimal offer schedule for a VRFB participating in ERCOT’s fre-
quency regulation market. The results of the optimization provide
valuable insight about how a VRFB might operate for regulation
service, and the value of that service in ERCOT. As an example,
we demonstrate the results of the optimization for a 1-h VRFB
operating on July 19th, 2009 in ERCOT’s southern hub. Fig. 9 shows
the inputs to the model for this day, and Fig. 10 shows the optimal
value of the decision variables computed from those inputs. Fig. 11
illustrates the dynamic state of the nominal, 1-h VRFB based on the
value of the decision variables in Fig. 10 and the VRFB dispatch sig-
nal in Fig. 9.
ð21Þ We obtain results like those of Figs. 10 and 11 for each day of
2007–2009. We adjust the results to account for energy losses as
in Eq. (13) and use Eq. (20) to reveal the revenue a VRFB could gen-
erate each day of 2007–2009 for regulation service in ERCOT. Illus-
trating our results, Fig. 12 shows the daily revenue a 1 MW VRFB
with a 1-h discharge duration could generate from regulation ser-
vice in ERCOT’s southern hub from 2007–2009. To show the poten-
tial value of a VRFB over the year in each region of ERCOT, we sum
the daily revenue from regulation service over 2007–2009 for each
of ERCOT’s four hubs. A useful figure for quantifying the value of
energy storage is its annual unit benefit: the revenue it generates
over a year divided by its rated AC power. Fig. 13 shows the annual
unit benefit of a 1-h VRFB operating in ERCOT’s four hubs for fre-

0.0
quency regulation from 2007–2009. Unlike locational energy
prices, regulation capacity prices are uniform throughout the ER-
COT region. Our results reflect this fact, showing that the location
−0.5 of a VRFB has little effect on its value for regulation service.
We use the results of our optimization routine to show how
−1.0
varying the electrolyte capacity of a VRFB affects its value for reg-
ulation service. Fig. 14 shows the mean annual unit benefit of
−1.0 −0.5 0.0 0.5 1.0
VRFBs ranging from 30 min to 4 h in discharge duration. Using
Pbatt Prated
our results, we estimate the lifecycle value of a VRFB system while
Fig. 8. We approximate VRFB energy losses from auxiliary plant equipment and the accounting for the time value of money. Given an anticipated
AC/DC inverter using a linear function. annual revenue stream, R, the present worth is estimated as in
 R.L. Fares et al. / Applied Energy 113 (2014) 189–198 195

URS Capacity Price VRFB Regulation Power Dispatched

30 1.0
25
0.5
($/MW)

20

(MW)
15 0.0
10
−0.5
5
0 −1.0
1 24 0 5 10 15 20 24

DRS Capacity Price VRFB Cell Current Density

14 60
12

(mA/cm2)
($/MW)

10
8 0
6
4
2
0 −60
1 24 0 5 10 15 20 24

Wholesale Energy Price VRFB Cell Voltage

50
40 1.50
($/MWh)

30 1.45

(V)
20 1.40
10 1.35
0 1.30
1 24 0 5 10 15 20 24

VRFB Dispatch Signal VRFB State of Charge

+1 URS −1 DRS

1.0 80
0.5 60

(%)
0.0
−0.5 40
−1.0 20
0 5 10 15 20 24 0 5 10 15 20 24
Hour Hour

Fig. 9. For each day of the year, the optimization problem for economic VRFB Fig. 11. Based on the offer schedule of power for frequency regulation in ERCOT, the
management takes in relevant prices from ERCOT, and the VRFB frequency internalized VRFB model dynamically estimates SOC by tracking the voltage and
regulation dispatch signal. The inputs for July 19th, 2009 are shown here. current.

Optimal URS Capacity Offer

4500
Daily Regulation Service Revenue ($)

1.0
(MW)

0.5 3000

0.0
1 24 1500

Optimal DRS Capacity Offer

1.0

0
(MW)

January 2007

July 2007

January 2008

July 2008

January 2009

July 2009

December 2009

0.5

0.0
1 24
Hour Fig. 12. From the results of the optimization, we estimate the revenue a 1 h, 1 MW
VRFB could generate from regulation service in ERCOT each day of 2007–2009.
Fig. 10. Based on the inputs in Fig. 9, the optimization routine computed the Results from ERCOT South are shown here.
optimal hourly URS and DRS capacity offers for the VRFB, shown here.

Eq. (23), where e is the annual price/cost escalation, d is the annual
discount rate, and L is the lifetime of the VRFB plant in years.
X L
ð1 þ eÞi0:5
PW ¼ R
i0:5
i¼1 ð1 þ dÞ
We use the standard assumption of a 2.5% per year annual price
escalation, 10% per year discount rate, and a nominal VRFB system
lifetime of 10 years [12,13,30]. Based on these assumptions, we ar-
rive on the present worth factor of the following equation:
PW ¼ 7:17 R ð24Þ
We use this present worth factor with the results of Fig. 14 to
show how the net present value (NPV) of different VRFB systems
ð23Þ used for frequency regulation service in Texas varies with system
cost. Fig. 15 shows this relationship for VRFB systems ranging from
30 min to 4 h in duration based on an annual revenue stream, R,
equal to the average annual unit benefit across the four regions
of Texas. Given an estimate of VRFB system cost, Fig. 15 could be
used to approximate the NPV of using the VRFB for frequency reg-
ulation alone.
196 R.L. Fares et al. / Applied Energy 113 (2014) 189–198

400
Table 3
North
South The U.S. Department of Energy Energy Storage Systems Program has estimated the
Annual Unit Benefit ($/kWAC−year)

Houston cost of different VRFB systems with varying annual charge/discharge duty cycles.
West Adapted from [30].
300
VRFB discharge duration (h) Cycles per year Estimated cost ($/kW)
1 20 1000
1 1000 1460
200 4 20 2701
4 250 3279

100

Table 4
Jossen and Sauer estimate the cost of power-related and energy-
related VRFB system components. Adapted from [33].
0
2007 2008 2009 Mean
Cost type Estimated cost
Fig. 13. The annual unit benefit of a 1-h VRFB used for frequency regulation in Power-related €1157/kW ($1511/kW)a
ERCOT varies significantly from year to year, ranging from $127–358/kW. Energy-related €78/kW h ($102/kW h)a
a
Based on 2013 exchange rate of $1.31 = €1.00.

300
North
Mean Annual Unit Benefit ($/kWAC−year)

South
250 Houston
West Table 5
Using our results and VRFB cost estimates from the U.S. Department of Energy [30],
200 we estimate the NPV of a 1-h and 4-h VRFB used for regulation service in ERCOT.

Discharge duration (h) Cost estimate ($/kW) NPV ($/kW)

150 1 1460–1613 (100)–53
4 1919–3279 (1408)–(48)

100

be affected by new vanadium mines and recycling of vanadium

50
waste products from steel slags, spent catalysts, and fly ash [32].
We utilize two separate cost estimates for a VRFB. The U.S.
0 Department of Energy’s (DOE) Energy Storage Systems Program
30 min 1 hr 2 hr 3 hr 4 hr
has estimated the cost of VRFB systems with various discharge
VRFB Discharge Duration
durations and frequencies of use [30]. Their cost estimates are
Fig. 14. The mean annual unit benefit of a VRFB from 2007–2009 increases with shown in Table 3.
increasing energy capacity. Jossen and Sauer performed a cost analysis that estimates the
cost of each major VRFB component [33,34]. VRFB components
can be classified as power-related components (e.g. activated car-
1000
30 minutes bon-felt electrode, bipolar current collector, ion-exchange mem-
1 hour brane) and energy-related components (e.g. V2O5 solute,
Net Present Value ($/kWAC)

2 hours
500 3 hours
electrolyte manufacture, tanks). Table 4 shows the power-related
4 hours and energy-related VRFB costs estimated by Jossen and Sauer.
Table 5 approximates the NPV of different VRFB systems based
0 on the cost estimates in Tables 3 and 4. Our results show that a 1-h
VRFB system could have a positive NPV if used for frequency reg-
ulation service in ERCOT. This being said, it is clear from the differ-
(500)
ing values in Tables 3 and 4 that there is uncertainty in the cost of a
VRFB. Thus, the results in Table 5 are not definitive and could
(1000) change with future energy storage cost estimates.
500 1000 1500 2000 2500

VRFB System Cost ($/kWAC) 5. Conclusion

Fig. 15. We estimate the net present value of various VRFB systems used for
frequency regulation in Texas based on economic optimization results from 2007– With the emergence of organized, competitive electricity mar-
2009. kets, and the rapid development of emerging electrochemical en-
ergy storage technologies, there has been renewed interest in the
business case for grid-connected battery energy storage. To show
It is difficult to estimate the cost of a commercial VRFB system the potential value of a novel battery for frequency regulation in
at this time. The cost of a VRFB is sensitive to ion-exchange mem- ERCOT, we implemented a control-oriented dynamic battery mod-
brane costs, electrode current density, electrode costs, and vana- el to describe the instantaneous energy storage capabilities of a
dium material costs [31]. Furthermore, it is difficult to gauge the large-scale VRFB. We used this model to perform a time-domain
long-term price of vanadium. VRFB commercialization efforts in analysis of a VRFB providing frequency regulation service in
China, Europe, and the United States led to price instability for ERCOT, utilizing an optimization routine to show how a VRFB could
vanadium oxide products in 2007 [32]. In the future, prices could operate. We used prices from ERCOT’s four hubs with the
 R.L. Fares et al. / Applied Energy 113 (2014) 189–198
optimization routine to show the annual unit benefit of various
VRFB systems operating in different regions of Texas for regulation
service. Based on our results, the present worth of a VRFB for fre-
quency regulation varies from approximately $1211/kW to
$1871/kW, depending on the system’s discharge duration. The lit-
erature approximates the cost of a 1-h VRFB system ranges from
$1460/kW to $1613/kW, and the cost of a 4-h VRFB ranges from
$1919/kW to $3279/kW [30,33]. While there is some uncertainty
in the cost of a VRFB, our results show that the additional cost of
liquid electrolyte is not offset by the added benefit of greater stor-
age capacity, causing a 1-h VRFB to have a greater NPV than a 4-h
VRFB used for regulation service in Texas. It should be noted that
our results are for a VRFB unit used for frequency regulation ser-
vice alone. A VRFB used for a combination of applications could
be more valuable [17], and have different tradeoffs regarding addi-
tional electrolyte/energy capacity.
In developing our estimates, we utilized the ERCOT automatic
generation control signal from 2007–2009, which reflects the
real-time difference between electric supply and demand on Texas’
electric grid. We assumed that a VRFB is dispatched in the direction
of the aggregate regulation power deployed on the grid. Because a
VRFB can provide on-demand power in milliseconds to seconds, a
VRFB could respond to deviations in frequency more quickly than a
traditional generator [22]. Thus, ERCOT could dispatch a VRFB
more rapidly than we assumed in this study, lessening frequency
deviations and reducing the total amount of regulation resources
procured [27]. It is estimated that rapid-response regulation ser-
vice might provide twice the benefit of regulation from electric
generators [5,13]. These findings are especially significant in light
of the projected increase in regulation capacity requirements with
additional ERCOT wind capacity [35]. To encourage development of
fast-ramping regulation resources, the United States Federal En-
ergy Regulatory Commission (FERC) issued Order 755 on October
20, 2011 [36]. FERC’s Order 755 seeks to adjust wholesale energy
market rules to internalize the benefits of fast-ramping regulation
resources. Because our analysis assumes a VRFB does not alter the
regulation dispatch schedule, we did not model how a VRFB might
affect overall regulation procurement. The relative benefit of fast-
ramping energy storage versus traditional electric generation for
regulation service is an interesting research question worthy of fu-
ture study.
While we have shown that a 1-h VRFB could have a positive
NPV if used for regulation service in ERCOT, we have not estab-
lished a control framework for a VRFB to realize the benefits quan-
tified in this paper. In the future, our work could be used to develop
a detailed model-predictive control framework for operational
management of a VRFB over diverse time scales. ERCOT has fore-
casting tools in place such as a day-ahead energy and ancillary ser-
vices market, and hour-ahead forecasting of real-time prices
[25,37]. Past work has shown that conventional price-forecasting
methods permit compressed-air energy storage to obtain approxi-
mately 95% of perfect-knowledge operating profits [26]. A model-
predictive controller could use price forecasts with a robust VRFB
model to manage battery operation in real time, while correcting
for model error.
Our work shows that large-scale electrochemical energy storage
devices such as RFBs could become cost-competitive with tradi-
tional frequency regulation technologies in the future as storage
technologies mature, and their cost decreases. Intelligent interpre-
tation of price signals for real-time battery control is an important
step in making grid-based batteries economically viable. We have
shown how a battery’s state might be monitored in an economic
framework, bridging the gap between high-level economic analysis
and future implementation of real-time battery control for maxi-
mum utility.
197
Acknowledgements
This work was sponsored by Pecan Street Inc. of Austin, Texas as
part of their ongoing smart grid demonstration project, which is
part of the American Recovery and Reinvestment Act Smart Grid
Demonstration Program administered by the U.S. Department of
Energy. Special thanks to Dr. Ross Baldick and ERCOT for facilitat-
ing access to the grid data utilized in this work.
References
[1] U.S. Energy Information Administration. Electricity; 2012. <ww.eia.gov/
electricity>.
[2] Electric Advisory Committee, U.S. Department of Energy. Energy storage
activities in the United States electric grid, tech. rep., U.S. Department of
Energy; 2011.
[3] Yang C-J, Jackson RB. Opportunities and barriers to pumped-hydro energy
storage in the united states. Renew Sustain Energy Rev 2011;15(1):839–44.
http://dx.doi.org/10.1016/j.rser.2010.09.020.
[4] Lachs W, Sutanto D. Application of battery energy storage in power systems.
In: Proceedings of International Conference on Power Electronics and Drive
Systems, vol. 2; 1995. p. 700–5. http://dx.doi.org/10.1109/PEDS.1995.404984.
[5] Eyer J. Benefits from flywheel energy storage for area regulation in California –
demonstration results. Tech. rep. SAND2009-6457, Sandia National
Laboratories; 2009.
[6] Makarov YV, Ma J, Lu S, Nguyen TB. Assessing the value of regulation resources
based on their time response characteristics. Tech. rep. PNNL-17632, Pacific
Northwest National Laboratory; 2008.
[7] Walawalkar R, Apt J, Mancini R. Economics of electric energy storage for energy
arbitrage and regulation in New York. Energy Policy 2007;35(4):2558–68.
doi:10.1016/j.enpol.2006.09.005.
[8] Anderson M, Carr D. Battery energy storage technologies. Proc IEEE
1993;81(3):475–9. doi:10.1109/5.241482.
[9] Sobieski D, Bhavaraju M. An economic assessment of battery storage in electric
utility systems. IEEE Trans Power Apparat Syst PAS-104 1985(12):3453–9.
doi:10.1109/TPAS.1985.318895.
[10] Butler PC, Iannucci J, Eyer J. Innovative business cases for energy storage in a
restructured electricity marketplace. Tech. rep., Sandia National Laboratories;
2003.
[11] Gyuk I, Kulkarni P, Sayer J, Boyes J, Corey G, Peek G. The united states of
storage. IEEE Power Energy Mag 2005;3(2):31–9. doi:10.1109/
MPAE.2005.1405868.
[12] Eyer J, Norris B. Guide to estimating benefits and market potential for
electricity storage in New York. Tech. rep., New York State Energy Research
and Development Authority; 2007.
[13] Eyer J, Corey G. Energy storage for the electricity grid: Benefits and market
potential assessment guide. Tech. rep. SAND2010-0815, Sandia National
Laboratories; 2010.
[14] Oudalov A, Chartouni D, Ohler C, Linhofer G. Value analysis of battery energy
storage applications in power systems. In: Power systems conference and
exposition, 2006., IEEE; 2006. p. 2206–11. doi:10.1109/PSCE.2006.296284.
[15] Oudalov A, Chartouni D, Ohler C. Optimizing a battery energy storage system
for primary frequency control. IEEE Trans Power Syst 2007;22(3):1259–66.
http://dx.doi.org/10.1109/TPWRS.2007.901459.
[16] Sioshansi R, Denholm P, Jenkin T, Weiss J. Estimating the value of electricity
storage in PJM: arbitrage and some welfare effects. Energy Econom
2009;31(2):269–77. doi:10.1016/j.eneco.2008.10.005.
[17] He X, Delarue E, D’haeseleer W, Glachant J-M. A novel business model for
aggregating the values of electricity storage. Energy Policy
2011;39(3):1575–85. doi:10.1016/j.enpol.2010.12.033.
[18] Hittinger E, Whitacre J, Apt J. What properties of grid energy storage are most
valuable? J Power Sources 2012;206(0):436–49. doi:10.1016/
j.jpowsour.2011.12.003.
[19] Linden D, Reddy TB. Handbook of batteries. McGraw-Hill Professional; 2001.
[20] Manwell JF, McGowan JG. Lead acid battery storage model for hybrid energy
systems. Solar Energy 1993;50(5):399–405. doi:10.1016/0038-
092X(93)90060-2.
[21] Zhao P, Zhang H, Zhou H, Chen J, Gao S, Yi B. Characteristics and performance
of 10 kW class all-vanadium redox-flow battery stack. Journal of Power
Sources 2006;162(2):1416–20. doi:10.1016/j.jpowsour.2006.08.016.
[22] EPRI, DOE, EPRI-DOE handbook of energy storage for transmission and
distribution applications. Tech. rep., Electric Power Research Institute; 2003.
[23] Bindner H, Krog Ekman C, Gehrke O, Isleifsson F. Characterization of vanadium
flow battery. Tech. rep., Danmarks Tekniske Universitet, Risø
Nationallaboratoriet for Bæredygtig Energi; 2010.
[24] Ma X, Zhang H, Sun C, Zou Y, Zhang T. An optimal strategy of electrolyte flow
rate for vanadium redox flow battery. J Power Sources 2012;203(0):153–8.
doi:10.1016/j.jpowsour.2011.11.036.
[25] ERCOT. Market information; 2012. <http://www.ercot.com/mktinfo/>.
[26] Townsend AK. A grid-level assessment of compressed air energy storage in
ERCOT. Ph.D. thesis, The University of Texas at Austin; 2013.
[27] Texas Energy Storage Alliance. Storage participation in ERCOT. White paper,
198 R.L. Fares et al. / Applied Energy 113 (2014) 189–198
Texas Energy Storage Alliance; 2010.
[28] NERC Resources Subcommittee. Balancing and frequency control. Tech. rep.,
North American Energy Electric Reliability Corporation; 2009.
[29] ERCOT Protocols. Electric Reliability Council of Texas; 2010.
[30] Schoenung S. Energy storage systems cost update. Tech. rep. SAND2011-2730,
Sandia National Laboratories; 2011.
[31] Zhang M, Moore M, Watson JS, Zawodzinski Ta, Counce RM. Capital cost
sensitivity analysis of an all-vanadium redox-flow battery. J Electrochem Soc
2012;159(8):A1183–8. doi:10.1149/2.041208jes.
[32] Skyllas-Kazacos M, Chakrabarti MH, Hajimolana SA, Mjalli FS, Saleem M.
Progress in flow battery research and development. J Electrochem Soc
2011;158(8):R55–79. doi:10.1149/1.3599565.
[33] Jossen A, Sauer DU. Advances in redox-flow batteries. In: First international
renewable energy storage conference; 2006.
[34] Kear G, Shah AA, Walsh FC. Development of the all-vanadium redox flow
battery for energy storage: a review of technological, financial and policy
aspects. Int J Energy Res 2012;36(11):1105–20. doi:10.1002/er.1863.
[35] Walling RA. GE energy, analysis of wind generation impact on ERCOT ancillary
services requirements. Tech. rep., General Electric Energy; 2008.
[36] Federal Energy Regulatory Commission. Order no. 755, Docket Nos. RM11-7-
000 and AD10-11-000.
[37] ERCOT. ERCOT launches wholesale pricing forecast tool. Press Release; July
2012.