Energy Policy 80 (2015) 7687

Contents lists available at ScienceDirect

Energy Policy
journal homepage: www.elsevier.com/locate/enpol

Resource and revenue potential of California residential load

participation in ancillary services$
Johanna L. Mathieu a,n, Mark E.H. Dyson b, Duncan S. Callaway b
a
b

Department of Electrical Engineering and Computer Science, University of Michigan, 1301 Beal Ave, Ann Arbor, MI 48109, USA
Energy and Resources Group, University of California at Berkeley, 310 Barrows Hall, Berkeley, CA 94720-3050, USA

H I G H L I G H T S

California's energy storage mandate requires 1.325 GW of energy storage by 2020.

Residential loads such as refrigerators have thermal energy storage.
California's residential loads could provide 10-40 GW/8-12 GWh of storage.
Loads participating in ancillary services markets could earn up to $56/load/year.
Consumer choices and policy mechanisms could increase revenue potentials.

art ic l e i nf o

a b s t r a c t

Article history:
Received 28 August 2014
Received in revised form
16 December 2014
Accepted 24 January 2015
Available online 5 February 2015

Increasing penetrations of intermittent renewable energy resources will require additional power system
services. California recently adopted an energy storage mandate to support its renewable portfolio
standard, which requires 33% of delivered energy from renewables by 2020. The objective of this paper is
to estimate the amount of energy storage that could be provided by residential thermostatically controlled loads, such as refrigerators and air conditioners, and the amount of revenue that could be earned
by loads participating in ancillary services markets. We model load aggregations as virtual energy storage, and use simple dynamical system models and publicly available data to generate our resource and
revenue estimates. We nd that the resource potential is large: 1040 GW/812 GWh, which is signicantly more than that required by the mandate. We also nd that regulation and spinning/nonspinning reserve revenues vary signicantly depending upon type of load and, for heat pumps and air
conditioners, climate zone. For example, mean regulation revenues for refrigerators are $11/year, for
electric water heaters are $24/year, for air conditioners are $0-32/year, and for heat pumps are $2256/
year. Both consumer choices, such as appliance settings, and policy, such as the design of ancillary service
compensation and appliance standards, could increase revenue potentials.
& 2015 Elsevier Ltd. All rights reserved.

Keywords:
Demand response
Energy storage
Ancillary services
Technical potential
Regulation
Appliances

1. Introduction

Abbreviation: AC, air conditioner; CAISO, California Independent System Operator; CEC, California Energy Commission; EIA, energy information agency; EWH,
electric water heater; GH, gas heater; GWH, gas water heater; HP, heat pump;
LADWP, Los Angeles Department of Water and Power; MAEC, mean annual energy
consumption; PG&E, Pacic Gas and Electric Company; RECS, residential energy
consumption survey; RF, refrigerator; SCE, Southern California Edison Company;
SDG&E, San Diego Gas & Electric Company; SMUD, Sacramento Municipal Utility
District; TCL, thermostatically controlled load

This work was funded, in part, by a Power System Engineering Research Center
(PSERC) Future Grid Initiative Grant.
n
Corresponding author. Tel.: 1 734 936 3875.
E-mail addresses: jlmath@umich.edu (J.L. Mathieu),
mehd@berkeley.edu (M.E.H. Dyson), dcal@berkeley.edu (D.S. Callaway).
http://dx.doi.org/10.1016/j.enpol.2015.01.033
0301-4215/& 2015 Elsevier Ltd. All rights reserved.

To reduce greenhouse gas emissions many states have implemented renewable portfolio standards that require a certain
percentage of electricity to come from renewable sources. Both
wind and solar photovoltaics are expected to comprise a signicant portion of new renewables (Loutan et al., 2007); however,
both technologies produce intermittent and uncertain power. As a
result, system operators will need to procure more ancillary services such as regulation and load following (Makarov et al., 2009;
Halamay et al., 2011). Rather than using power plants to provide
these additional services, it may be more cost-effective and/or
environmentally benecial to use alternative technologies, namely
energy storage devices (e.g., batteries, ywheels, compressed gas,

J.L. Mathieu et al. / Energy Policy 80 (2015) 7687

and pumped hydropower) and demand response resources.

In October 2013, California became the rst state in the U.S.
with an energy storage mandate (Sweet, 2013). The mandate requires 1325 MW of energy storage by 2020 (CPUC, 2013a), which is
expected to support California's renewable portfolio standard goal
of 33% of delivered energy from renewables by 2020 (CPUC,
2013b). California's denition of energy storage includes systems
that store thermal energy for direct use for heating or cooling at a
later time in a manner that avoids the need to use electricity at
that later time (State of California, 2010). Therefore, demand response that shifts the power consumption of thermostatically
controlled loads (TCLs), such as heating and air conditioning systems, electric water heaters, and refrigerators, which all store heat
in thermal mass, may qualify as energy storage, and contribute to
California's energy storage needs.
Residential TCLs are well-suited to load shifting on timescales
of seconds to minutes. These loads generally operate with hysteresis, modulating temperature between an upper and lower
limit, and so their power consumption is inherently exible. Recent work has focused on the development of strategies to coordinate the power consumption of large aggregations of residential TCLs to provide power system services, e.g., (Callaway,
2009; Kundu et al., 2011; Perfumo et al., 2012; Bashash and Fathy,
2013; Mathieu et al., 2013b; Zhang et al., 2013; Totu et al., 2013),
while ensuring that coordination actions are non-disruptive to
electricity consumers, meaning that temperatures stay near or
within existing temperature limits (Callaway and Hiskens, 2011). A
variety of approaches have been shown effective in simulation;
however, several key questions remain including: How well will
loads perform in practice? What is the size of this resource? Is it
large enough to play a meaningful role in intra-hour energy balancing? Will revenue earned from ancillary services participation
be enough to cover the infrastructure and operational costs associated with non-disruptive load coordination? While answering
the rst question requires testbeds and pilot studies, we can begin
to develop answers to the last three questions with data and
models.
The objective of this paper is to develop order-of-magnitude
estimates for the technical resource potential and ancillary service
revenue potential of non-disruptive coordination of residential
TCLs in California. To develop these estimates, we model TCL aggregations as virtual energy storage devices and rst estimate the
time-varying and location-dependent power and energy capacities
of aggregations of four types of residential TCLs central air
conditioners, heat pump heating systems, electric water heaters,
and refrigerators. We then generate statewide technical resource
potential estimates (i.e., estimates of the time-varying aggregate
power and energy capacity of all TCLs in California) for both 2014
and 2020, where in the latter we consider two scenarios:
(i) electric appliance saturation levels equivalent to those today
and (ii) increased electric appliance saturation levels due to a
portion of consumers switching from gas to electric water and
space heating. The latter analysis is not meant to be predictive but
rather to explore the effect of increased appliance electrication as
recommended by a California Council on Science and Technology
report (CCST, 2011), which listed appliance electrication as one of
the changes needed to achieve greenhouse gas emissions reductions to 80% below 1990 levels by 2050. With our resource estimates, we estimate the mean revenue per TCL per year for TCLs
providing regulation and spinning/non-spinning reserve. This paper builds upon and updates our preliminary work in this area
(Mathieu et al., 2012).
Our contributions are twofold. First, we develop methods to
estimate resource and revenue potentials of TCL aggregations.
These methods rely on simple models of TCL dynamics and publicly available data, and could be used to generate estimates for

77

other regions. Second, our estimates inform energy policy. An

understanding of the size of the resource and current nancial
incentives will help policy makers determine whether policy
mechanisms are needed to achieve load participation in ancillary
services markets.
This paper is organized as follows. In Section 2 we present our
methods including appliance ownership, temperature, and price
data sources; TCL model; technical resource potential calculation
methods; and revenue potential calculation methods. Section 3
presents the results of our resource and revenues potential analyses, and Section 4 provides a discussion of these results in the
context of energy policy. Section 5 gives concluding remarks.

2. Methods
2.1. Appliance ownership, temperature, and price data sources
We estimated the number of central air conditioners (henceforth air conditioners and abbreviated AC), heat pump heating
systems (henceforth heat pumps and abbreviated HP), refrigerators (RF), electric water heaters (EWH), gas heating systems
(henceforth gas heaters and abbreviated GH), and gas water
heaters (GWH) in ve utility districts in California: Pacic Gas and
Electric (PG&E), Southern California Edison (SCE), San Diego Gas
and Electric (SDG&E), Sacramento Municipal Utility District
(SMUD), and Los Angeles Department of Water and Power
(LADWP). We did not consider room air conditioners or space
heaters because the resource is small and hard to control. To develop the estimates, we used appliance saturation rates from the
2009 California Residential Appliance Saturation Survey (Palmgren
et al., 2010) and we extrapolated California Energy Commission
(CEC) forecasts for the number of households in California in 2014
and 2020 (Marshall and Gorin, 2007). For air conditioners, heat
pumps, and gas heaters, we used saturation rates by CEC Forecasting Climate Zone (listed in Table 1 and shown in Fig. 1a) since
power consumption varies as a function of outdoor air temperature. For refrigerators, electric water heaters, gas water heaters, we
used statewide saturation rates since power consumption varies as
a function of indoor air temperature, which we assume is constant
and uniform across the state. Table 2 shows estimates for the
number of households per CEC Forecasting Climate Zone in 2014
and 2020 together with appliance saturation rates. Note that
newer statewide forecasts are available (Kavalec and Gorin, 2009;
Alcorn et al., 2013a, 2013b); however, these reports do not provide
breakdowns of households by climate zone. The 2007 statewide
household forecasts (13.14 million in 2014, 14.26 million in 2020)
Table 1
List of CEC Forecasting Climate Zones, mapping to California Building Climate
Zones, and Associated CAISO Ancillary Service Zone.
CEC Forecasting Climate
Zone

California Building Climate Zone

CAISO Ancillary Service Zone

1: PG&E North Coast

Mountain
2: PG&E Sacramento Area
3: PG&E Central Valley
4: PG&E East Bay
5: PG&E San Francisco
6: SMUD
7: SCE San Joaquin
8: SCE LA Basin Coast
9: SCE LA Basin Inland
10: SCE Inland Empire
11: LADWP
12: LADWP
13: SDG&E

1: Arcata

North

12: Sacramento
12: Sacramento
2: Santa Rosa
3: Oakland
12: Sacramento
13: Fresno
6: Los Angeles
10: Riverside
14: China Lake
10: Riverside
10: Riverside
7: San Diego

North
North
North
North
North
South
South
South
South
South
South
South

78

J.L. Mathieu et al. / Energy Policy 80 (2015) 7687

Fig. 1. California climate zones. Source: CEC (reprinted with permission). (a) CEC Forecasting Climate Zones (KEMA-XENERGY, 2004). (b) California Building Climate Zones
(CEC, 2011).
Table 2
Numbers of households (extrapolated from Marshall and Gorin, 2007) and
appliance saturation rates (Palmgren et al., 2010) by CEC Forecasting Climate Zone.
Climate zone Households
(millions)

1
2
3
4
5
6a
7
8
9
10
11 & 12b
13
Total

Appliance saturation rates

2014

2020

AC

HP

0.30
0.48
1.22
1.78
1.42
0.56
0.23
2.20
0.90
1.40
1.35
1.30
13.14

0.35
0.59
1.40
1.90
1.45
0.61
0.26
2.30
1.00
1.60
1.35
1.45
14.26

0.43
0.79
0.77
0.46
0.10
0.79
0.61
0.44
0.54
0.80
0.41
0.43

0.01
0.00
0.02
0.01
0.02
0.00
0.00
0.02
0.01
0.01
0.00
0.02

RF

EWH GH

GWH

0.31
0.73
0.77
0.83
0.81
0.73
0.67
0.82
0.80
0.83
0.81
0.73
1.25

0.07

0.74

Independent System Operator (CAISO) Open Access Same-Time

Information System (OASIS) (CAISO, 2014b). We obtained hourly
prices for up and down regulation,1 spinning reserve, and nonspinning reserve from January 1, 2010 to November 30, 2014 (with
the exception of the 2012 leap day) for both of CAISO's ancillary
service zones North and South. We converted all prices into 2014
dollars (USD 2014). We did not use prices from 2009 or earlier
since CAISO underwent a signicant market redesign in 2009
(CPUC, 2008). We computed revenue potentials for air conditioners and heat pumps by climate zone using prices associated
with the correct ancillary service zone. While most climate zones
are clearly located within either the North or South ancillary service zone, some climate zones straddle the border. In those cases,
we assumed the climate zone was within the ancillary service
zone containing the majority of the land mass. In Table 1, we list
the ancillary service zones we associated with each climate zone.
2.2. TCL modeling

Saturation rates were unavailable for climate zone 6 (SMUD), and so we use
rates from climate zone 2 (PG&E Sacramento Area), which experiences a similar
climate.
b
We were unable to nd separate household forecasts for climate zones 11 and
12 and so we treat them as a single zone.

are close to the 2013 forecasts (12.813.0 million in 2014, 13.4

14.2 million in 2020).
We used typical meteorological year outdoor air temperature
data from EnergyPlus (EERE, 2012). Since the California Building
Climate Zones used by EnergyPlus (Fig. 1b) are different than the
CEC Forecasting Climate Zones (Fig. 1a), we mapped each CEC
Forecasting Climate Zone to a California Building Climate Zone, as
shown in Table 1.
We used ancillary services prices from the California

TCLs operate within temperature hysteresis bands (commonly

called dead-bands in the TCL literature), as shown in Fig. 2.
Specically, a thermostat switches a TCL ON and OFF to maintain
an internal temperature within the temperature band [ , + ],
where = s /2, + = s + /2, s is the temperature set point,
and is the temperature dead-band width. When a TCL is ON, it
consumes a xed amount of power P and the mode q is 1. When a
TCL is OFF is consumes no power and the mode q is 0. We use the
discrete-time model developed by Chong and Debs (1979) and
Ihara and Schweppe (1981) to describe the dynamics of a cooling
TCL i:
1
We only use capacity prices, not mileage prices, since the mileage market is still
in development.

J.L. Mathieu et al. / Energy Policy 80 (2015) 7687

79

Let hpi be the amount of time it takes TCL i to traverse its deadband in ON mode. Then, P ihpi is the amount of energy required by
TCL i to traverse its dead-band in ON mode. A TCL aggregation's
energy capacity Ecap is the total amount of energy needed by all
available TCLs to traverse their dead-bands in ON mode minus the
mean energy required over hpi under normal operation. It is computed as (Mathieu et al., 2013a, 2014a)

state

ON

OFF

Na
-

Ecap =

temperature

(4)

i=1

Fig. 2. The temperature dead-band of a cooling TCL.

ki + 1 = ai ki + (1 ai)( a,i k qki gi ),

where Na is the number of available TCLs and D is a TCL's duty

cycle. If we assume a is xed, hp and D can be computed from (1),
specically

(1)

hpi = RiC i ln

0, i
i
k + 1 <

i
i
qk + 1 = 1, k + 1 > +i ,

qk , otherwise

i ai + gi
+i ai + gi

,
(5)

and

(2)

yki = qki P i,

Di =

hpi
hpi + hui

,
(6)

(3)

where k is the time index, a is the ambient temperature (i.e., the

temperature of the space surrounding the conditioned space), and
y is the TCL's instantaneous power consumption. The non-di(h / RiC i)

mensional parameter a is dened as: ai = e

, where h is the
model time step, C is a TCL's thermal capacitance, and R is its
thermal resistance. The temperature gain g is dened, for cooling
TCLs, as: gi = RiP ii , where

P ihpi (1 Di),

is the coefcient of performance. For

heating TCLs, the position of the 0 and 1 in (2) is switched and gi is

negative.
We assume the TCL model parameters are constants, though in
reality they would be time-varying. For example, would change
as a function of a (WECC, 2007). Additionally, this model does not
capture the full dynamics of a TCL. While more complicated
models are available from the literature, e.g., (Schneider et al.,
2011; Zhang et al., 2013; Vrettos et al., 2012), this model captures
the key characteristics of TCLs, specically their hybrid system
dynamics, thermal capacitance, and thermal resistance, and
therefore is sufciently accurate for order of magnitude estimates
of the resource/revenue potential of large aggregations of TCLs.
To ensure that load coordination actions are non-disruptive to
consumers, we assume that a load aggregator can switch TCLs ON
and OFF only when they are within their temperature dead-bands,
and we refer to such TCLs as available. TCLs outside of their deadbands are unavailable and do not respond to control signals. A
cooling TCL is available when would be larger than without
cooling and the TCL is capable of cooling the space to within the
dead-band, given its parameters and a . If a was xed, a cooling
TCL would be available when a > +i and a RiP ii < i . We can
compute similar conditions for heating TCLs. When a is not xed,
it is more difcult to determine TCL availability, as described in
Section 2.3.2.
2.3. Calculating the technical resource potential
Building upon our past work (Mathieu et al., 2012, 2013a,
2014a, 2014b), we propose a method to quantify the TCL resource
potential by representing TCL aggregations as virtual energy storage devices with an energy capacity and power capacity. The energy capacity of a TCL aggregation is analogous to the maximum
amount of energy that can be stored in a battery. The power capacity is analogous to the full power range of a battery, i.e., its
maximum injection power plus its maximum extraction power.

where hu is the amount of time it takes a TCL to traverse its deadband in OFF mode, which, assuming a is xed is

hui = RiC i ln

+i ai
i ai

Note that only positive values of

(7)
hpi

and

hui

are used to compute Di.

When hpi is negative, the TCL is unavailable and Di 1, and when hui
is negative, the TCL is also unavailable but Di 0.
The power capacity Pcap is dened as the sum of the maximum
power consumption of all available TCLs (Mathieu et al., 2013a,
2014a)
Na

Pcap =

P i.
i=1

(8)

We next detail our methods for computing the energy and

power capacities of aggregations of refrigerators (Section 2.3.1), air
conditioners and heat pumps (Section 2.3.2), and electric water
heaters (Section 2.3.3). Our calculations require accurate estimates
of all TCL parameters. In each case, we identify a set of parameters
that we can estimate relatively well based on published data and/
or simple models, and then we tune the remaining parameters by
matching the mean annual energy consumption (MAEC) predicted
by the model to actual MAECs computed with data from the 2009
U.S. Energy Information Agency (EIA) Residential Energy Consumption Survey (RECS) (EIA, 2009). Table 3 shows all tuned TCL
parameters and MAECs. We assume that parameters are drawn
from uniform distributions bounded by the values shown; uniform
distributions are appropriate in situations such as these where the
real underlying distribution's characteristics (for example the
mode or skewness) are unknown but a plausible, nite range of
the support is known (Morgan and Small, 1992). We refer to aggregations of loads as heterogeneous since each load has different parameters.
2.3.1. Refrigerators
The mean power consumption of an aggregation of TCLs is
N

Pm =

P iDi ,
i=1

(9)

where N is the number of TCLs. As stated in Section 2.1, we assume

indoor air temperature is constant and uniform across the state.

80

J.L. Mathieu et al. / Energy Policy 80 (2015) 7687

Table 3
TCL parameters and mean annual energy consumptions (MAECs).
Parameter

AC

RF

HP

EWH

Ambient temperature, a (C)

Dead-band width, (C)
Temperature set point, set (C)
Thermal resistance, R (C/kW)
Thermal capacitance, C (kWh/C)
Coefcient of performance, ()
Power consumption, P, (kW)
Model-predicted MAEC per TCL
(kWh)b
Actual MAEC per TCL (kWh)

Variable
0.251.0
1827
1.52.5a
1.52.5
23
47.2
1550

20
12
1.73.3
80105a
0.40.8
1.52.5
0.20.5a
850

Variable
0.251.0
1524
1.52.5
1.52.5
34
47.2
4780

20
24
4354
see Fig. 3a
0.20.6
12
4.5
1870

1569c

854d

N/A

1874e

Tuned parameters.
Numbers reported are the mean of 10 model runs.
c
The 2009 RECS (EIA, 2009) does not give MAECs separately for central and
room air conditioners, though the 2005 RECS (EIA, 2005) does. To compute the
MAEC for central air conditioners, we assume the ratio of energy consumption of
central to room air conditioners is as in 2005 (1:0.37), the ratio of ownership is as
in 2009 (1:0.4), and the overall MAEC is as in 2009 (1288 kWh).
d
The MAEC per refrigerator is computed with the MAEC of refrigerators per
household (EIA, 2009) and mean refrigerator saturation rate (Palmgren et al.,
2010).
e
From EIA (2009), which reports the MAEC for electric water heaters using any
electric water heating technology (e.g., electric resistance, heat pump).
b

With a xed, each refrigerator's duty cycle is constant, and so Pm ,

Pcap , and Ecap are constant. To compute the MAEC, we simply
multiply Pm by the number of hours in a year, i.e.,
MAEC = Pm 8760 h .
We estimated ranges for , set , C, and (Mathieu, 2012), and
then tuned the ranges of R and P until the model-predicted MAEC
approximately equaled the actual MAEC and the mean duty cycle
approximately equaled 33% (EERE, 2014). We then used the tuned
parameters to compute the energy and power capacities.
Note that, in reality, the duty cycle and the mean power consumption of an aggregation of refrigerators will change with
changing indoor air temperatures and disturbances including
door openings, warm casseroles put in the refrigerator to cool, and
frozen chickens put in the fridge to thaw. Here, we assume the
effect of disturbances is small and ignore them. Indoor air temperature a has no effect on power capacities and essentially no
effect on energy capacities (assuming a is not near the dead-band
temperatures). However, because a affects Pm , it also affects our
revenue estimates, as will be described in Section 2.4.
2.3.2. Air conditioners and heat pumps
It is not reasonable to assume that outdoor air temperature is
xed, and so, for air conditioners and heat pumps, we assume a
varies hourly. As a result, D, Pm , Pcap , and Ecap all change over time.
We rst use (9) to compute the mean power consumption of an
aggregation of heat pumps or air conditioners that all experience
the same (xed) a , and we repeat this calculation for all physically
possible values of a . This allows us to generate a lookup table
relating Pm to a . Then, for each hour of the year j, we can determine Pm, j based on the hourly outdoor temperature a, j . Since
consumer behavior affects air conditioner and heat pump usage,
we assume heat pumps are not available when the mean daily
temperature is greater than 15 C and air conditioners are not
available when the mean daily temperature is less than 20 C. On
those days, we set Pm, j = 0 for all hours j within the day. The MAEC
of an aggregation of heat pumps or air conditioners that all experience the same outdoor temperatures in a year is
8760

MAEC =

(Pm, j 1 h).

j=1

We assume that all heat pumps or air conditioners within a

climate zone experience the same outdoor temperatures. Then,

the statewide MAEC is the weighted mean of the MAECs associated with each climate zone, where the weights are proportional
to the number of heat pumps or air conditioners in each climate
zone.
For air conditioners, we estimated ranges for , set , C, P, and
(Mathieu, 2012), and tuned the range of R until the model-predicted MAEC approximately equaled the actual statewide MAEC.
RECS reports a MAEC for space heating, but the value is dominated
by the energy consumption of room space heaters rather than
central heat pump heating systems, and so we do not use it. Instead, we use the tuned air conditioner parameters, but increase
and decrease set to reect typical values for heat pumps (EIA,
2009). We used the tuned parameters to compute energy and
power capacities as a function of ambient temperature. Then, we
summed the energy and power capacities associated with each
climate zone to create the statewide estimates.
Our initial methodology (Mathieu et al., 2012) used the full
temperature dynamics for model tuning, which was likely more
accurate, but time-consuming. Our current approach, also used in
Mathieu et al. (2014b), effectively assumes a TCL's behavior is a
function of only the current value of a , not past values. This approximation is small if always remains near dead-band temperatures, including when neither heating nor cooling is used. If
this is the case, the thermal mass of the house also remains near
dead-band temperatures, and so the mean power consumption,
power capacity, and energy capacity of the TCL aggregation is
simply a function of the current a . However, if, for example, a
house cools down over a summer night, the following morning, as
a rises, the house will not need air conditioning until the thermal
mass of the house heats to the dead-band temperatures, which
will occur long after a passes the dead-band temperatures. In this
case, the mean power consumption, power capacity, and energy
capacity of the TCL aggregation is a function of both current and
past values of a .
2.3.3. Electric water heaters
Electric water heater power usage is driven by water consumption patterns. We assume that the portion of the daily energy
consumption that is consumed in each hour of the day is equivalent to the portion of the daily water consumption that is consumed in each hour of the day. We obtained a mean water draw
prole from Lutz et al. (2006), which reports data from eld studies. Since we were unable to obtain more detailed data, we assume every day exhibits the same prole, i.e., we neglect seasonal
variation. Seasonal variation would be driven by varying ground
water temperatures, but will be much less than typical diurnal
variability. In Fig. 3, we plot the portion of water consumed in each
hour of the day (top panel) and the assumed mean power consumption in each hour of the day (middle panel).
Water draws result in a temperature drop in the tank, which
can be modeled as a change in R. We estimate ranges for , set , C,
P, and (Mathieu, 2012), and tuned the range of R separately for
each hour of the day l such that Pm, l 1 h matches the mean energy consumed in that hour, calculated as l MAEC/365, where l
is the portion of the daily water/energy consumption that is consumed in hour l. To simplify our tuning process, we assumed
Rmax Rmin = 100, where R [Rmin , Rmax ]. Tuned ranges for R are
shown in Fig. 3 (bottom panel). We use (4) and (8) to compute
energy and power capacities for each hour of the day. Note that
this methodology differs from the approach in Mathieu et al.
(2012), which assumed R was constant throughout the day.
As with refrigerators, we assume a is xed. This assumption
will not affect our resource potential estimate, but will affect our
revenue estimates.

J.L. Mathieu et al. / Energy Policy 80 (2015) 7687

81

Water Consumption
%

0.2
0.1
0

12

14

16

18

20

22

24

16

18

20

22

24

16

18

20

22

24

Mean Power Consumption, Pm

1000
kW

10

500
0

10

12

14

Tuned ranges for R

C/kW

500

10

12
14
hour of day

Fig. 3. Hourly electric water heater water consumption as a percentage of the daily water consumption (Lutz et al., 2006), mean power consumption of 1000 heterogeneous
electric water heaters, and tuned ranges for R.

2.4. Calculating the revenue potential

We rst describe our assumptions with respect to load participation in ancillary services markets. We assume that load aggregators (which may be utilities or third-party companies) would
compete against all other market participants, including generators and storage, to provide ancillary services. As is done today,
each resource would bid into the ancillary services market with a
quantity/price curve and the market would clear with a uniform
market clearing price. Only resources with a lower price than the
market clearing price will be called upon to participate.
Assuming TCLs provide ancillary services at a lower cost than
other resources and assuming they comprise a small portion of the
total market (and therefore do not affect the market clearing
price), we can use historical market clearing prices to estimate
how much money they could earn by participating in ancillary
services markets. In actuality, we do not yet know the full costs
required to enable TCL participation in ancillary services. Therefore, our revenue potential estimates should not be interpreted as
predictive of actual revenues, but rather provide load aggregators
with a challenge if TCL ancillary service participation could be
enabled at lower costs that those report here, they would likely
make a prot given current market prices, participants, and
structure. Of course, in the future, any of these things may change
ancillary service prices may increase with higher penetrations of
variable renewable resources, energy storage costs may decrease,
and electricity market rules may evolve. All of these changes
would necessitate a new revenue potential analysis.
We compute the revenue potential associated with participation in
(1) regulation (also referred to as automatic generation control, load
frequency control, and secondary frequency control), in which resources follow dispatch signals on the timescale of seconds; and (2)
spinning/non-spinning reserve in which online/ofine resources are
dispatched on timescales of minutes to hours. In California, resources
participating in regulation must follow a signal that changes ever 4 s,
while resource participating in spinning/non-spinning reserve must
respond to dispatch signals within 10 min and be able to maintain the
response for at least two hours (CAISO, 2006). In California, loads can
provide regulation, spinning reserve, and non-spinning reserve via the
participating load functionality (CAISO, 2009).
To participate in an ancillary services market, resources must
offer a power capacity (in kW), and, if the offer is accepted, they
are dispatched within that capacity. We dene a TCL aggregation's
regulation power capacity Preg as its sustained 15-min capacity and
its spin/non-spin power capacity Psp/ns as its sustained 2-h capacity. We choose 15-minutes for regulation to be consistent with

CAISO's Regulation Energy Management functionality that enables

non-generator resources (NGRs) like storage and loads to participate in regulation by bidding their 15-min power capacity (CAISO,
2011, 2014a). The CAISO real-time market is used to manage the
state of charge of NGRs (CAISO, 2011). We choose 2 h for spinning/
non-spinning reserves because of CAISO's minimum response time
requirement.
To compute Preg and Psp/ns , we rst assume that, when not participating in regulation or spinning/non-spinning reserve, the
average TCL internal temperature is equivalent to the average TCL
set point. This means that the TCL aggregation could increase or
decrease its energy consumption by half of its energy capacity,
Ecap/2, and so its energy state (similar to the state of charge of a
battery) is 50%. We also assume, for simplicity, that a TCL aggregation must provide the same amount of up (below Pm ) and
down (above Pm ) capacity. With these assumptions, we compute a
TCL aggregation's regulation and spin/non-spin power capacities as

Preg = min (4(Ecap/2), Pcap Pm, Pm ),

(10)

Psp/ns = min ( (Ecap/2)/2, Pcap Pm, Pm ).

(11)

Revenues potentials are computed by multiplying the hourly

regulation or spin/non-spin power capacity by the hourly regulation or spinning/non-spinning reserve price. Since we assume that
TCL aggregations must provide the same amount of up and down
capacity, in each hour we sum CAISO's up and down regulation
price to obtain the total regulation price.
Regulation and spin/non-spin power capacities are functions of
Pm which is a function of a . For heat pumps and air conditioners,
we consider a time-varying a (outdoor air temperature), while for
refrigerators and electric water heaters, we assume a is constant.
A sensitivity analysis shows that a 2 C change in a results in an
11% change in revenues for refrigerators and a <1% change in
revenue for electric water heaters. In the following section, we
report mean revenues per year over all refrigerators and electric
water heaters and so, provided our choice of a is correct on
average, our estimates are reasonable.

3. Results
3.1. Resource potential
We report the estimated power and energy capacities of aggregations of 1000 heterogeneous TCLs, by TCL type, in

82

J.L. Mathieu et al. / Energy Policy 80 (2015) 7687

Tables 4 and 5. In the case of refrigerators, the power and energy

capacities are constant throughout the year because we assume
the TCL parameters and ambient temperature are constant. Despite allowing the thermal resistance of electric water heaters to
vary from hour to hour, we nd that the power and energy capacity of electric water heaters are approximately constant
through the year as well. This would not be the case if Rmin was
sufciently small that some electric water heaters were unable to
keep their water temperatures within their dead-bands (resulting
in a duty cycle of one and the electric water heater being unavailable). However, for the tuned ranges for R, shown in Fig. 3, all
TCLs remain available in all hours, resulting in a constant Pcap . Ecap
is also approximately constant because it is effectively insensitive
to the choice of R. However, as also shown in Fig. 3, Pm is not
constant throughout the day.
The power and energy capacities of aggregations of air conditioners and heat pumps are a function of outdoor air temperature.
Fig. 4 shows a graphical version of the look-up table that relates
outdoor air temperature a to D, Pcap , Pm , and Ecap . Looking at the
right side of the plot, one can see that as outdoor air temperature
increases more air conditioners become available. At very high
outdoor air temperatures Pcap and Ecap fall because some air conditioners are unable to cool the space to within the dead-band,
causing them to become unavailable. We can see the opposite on
the left side of the plot as the outdoor air temperature decreases
more heat pumps become available, and at very low outdoor air
temperatures Pcap and Ecap fall because some heat pumps are unable to heat the space to within the dead-band, causing them to
become unavailable.
Table 5 shows the estimated mean annual power and energy
capacities of TCL aggregations by climate zone along with the
number of hours per year (out of 8760) that the aggregation has a
non-zero power and energy capacity Nh . Note that for refrigerators
and electric water heaters Nh = 8760.
In Fig. 5, we plot the time-varying power and energy capacities
of residential TCLs in California in 2014 and 2020, where, for the
latter, we investigate two scenarios. The results plotted in Fig. 5b
differ from those in Fig. 5a because they use forecasts for the
number of households in 2020 and assume that all TCLs become
5% more efcient (i.e., is multiplied by 1.05). The results plotted
in Fig. 5c additionally assume that 10% of gas heaters are replaced
by heat pumps and 10% of gas water heaters are replaced by
electric water heaters. We refer to this scenario as the 2020
electrication scenario. Though hard to see in the plots, power
and energy capacities increase from a peak of 44.4 GW/11.6 GWh
in 2014 to a peak of 49.0 GW/12.0 GWh in 2020. With electrication, these values further increase to 53.8 GW/12.8 GWh.
3.2. Revenue potential
Tables 6 and 7 list estimated regulation and spin/non-spin
power capacities of aggregations of 1000 heterogeneous TCLs
along with estimated revenues per TCL per year, by TCL type. We
report mean regulation and spin/non-spin power capacities, where
the mean is taken over the year. Note that refrigerator aggregations have constant regulation and spin/non-spin power capacities
(i.e., Preg (j) = Preg j , Psp (j) = Psp j , and Pns (j) = Pns j ), while all
Table 4
Power and energy capacities of aggregations of 1000 heterogenous refrigerators
and electric water heaters.

Refrigerators
Electric water heaters

Pcap (kW)

Ecap (kWh)

352
4500

457
854

Table 5
Mean annual power and energy capacities of aggregations of 1000 heterogenous air
conditioners and heat pumps, by climate zone. Columns labeled Nh list the number
hours per year that the aggregation has a non-zero power/energy capacity.
Climate zone

1
2
3
4
5
6
7
8
9
10
11 & 12
13

Air conditioners

Heat pumps

Nh (h)

Pcap (kW)

Ecap (kWh)

Nh (h)

Pcap (kW)

Ecap (kWh)

0
1402
1402
764
61
1402
2902
948
2113
2539
2113
1460

0
646
646
364
24
646
1371
261
965
1241
965
415

0
60
60
34
2
60
126
24
89
114
89
38

7608
4340
4340
4827
5012
4,340
3785
3609
3189
4076
3189
2909

4651
2585
2585
2873
2993
2585
2242
2089
1808
2418
1808
1620

304
169
169
188
196
169
147
136
118
158
118
106

other TCL types have time-varying capacities (due to time-varying

a for air conditioners and heat pumps, and due to time-varying
ranges for R for electric water heaters). We also report revenue
statistics per TCL per year. For electric water heaters and refrigerators, we estimate the mean revenue per TCL in each year
(20102014) and in each ancillary service zone (North and South),
and we take the mean and standard deviation over the 10 estimates. For heat pumps and air conditioners, we estimate the mean
revenue per TCL in each year (20102014), and we take the mean
and standard deviation over the 5 estimates (recall that, for each
climate zone, we use prices corresponding to the associated ancillary services zone).

4. Discussion
Examining the results presented in the previous section, we
nd that the resource potential of California residential loads is
large, and well above that required by California's energy storage
mandate. The peak power capacity in 2014 (E40 GW) is the same
order of magnitude as the statewide peak consumption
(E60 GW). While our estimates may seem too large, it is important to keep in mind the denition of power capacity the sum
of the maximum power consumption of all available TCLs. Without
coordination, the power consumption of residential TCLs is never
this large because TCL operation is unsynchronized. The peak is
driven by air conditioners and, because of the low penetration of
heat pumps, winter power capacities (E 9 GW in 2014) are much
lower than summer power capacities, though still approximately
seven times more than the mandate. The energy capacity varies
less, from 8.4 to 11.6 GWh in 2014, because most of the energy
capacity is provided by refrigerators, which are always available.
California has more refrigerators than any other type of TCL;
however, unlike all other TCLs, their energy capacity is greater
than their power capacity resulting in an unequal effect on the
statewide power and energy capacities, as shown in Fig. 5.
While the results plotted in Figs. 5a and b show similar trends,
Fig. 5c shows the effect of increasing the number of electric water
heaters and heat pumps. The increase in electric water heaters
simply increases the year-round capacity, while the increase in
heat pumps increases the capacity during the winter. Shoulder
seasons still include periods when essentially no air conditioners
or heat pumps are needed, and so total yearly power and energy
capacity variation is similar. Fig. 5 shows that heat pumps are used
more hours in a year than air conditioners, resulting in higher
mean power and energy capacities. However, even with the

J.L. Mathieu et al. / Energy Policy 80 (2015) 7687

83

Mean Duty Cycle, D

1
0.5

10

10

kW

4000

Pcap

2000

Pm
10

10

40

50

20

30

40

50

30

40

50

Energy Capacity, Ecap

600
kWh

30

Power Capacity, Pcap & Mean Power Consumption, Pm

6000

20

400
200
0

10

10

20
Temperature ( C)

Fig. 4. Mean duty cycle, power capacity and mean power consumption, and energy capacity of 1000 heat pumps (thin lines) and 1000 air conditioners (thick lines) as a
function of outdoor temperature. Two vertical lines delineate the regions of (i) heating (left), (ii) heating and cooling (middle), and (iii) cooling (right).

assumed increase in heat pumps in the 2020 electrication scenario, the number of air conditioners would be much larger than
the number of heat pumps.
It is interesting to compare our results to those of a recent
study that used smart meter data to estimate the resource potential of residential air conditioners in PG&E territory (Dyson
et al., 2014). The paper reports an estimate of 3.23.8 GW for
short-term load curtailment. Given that PG&E's peak comprises
about 39% of California's peak demand (Alcorn et al., 2013b), we
could extrapolate this estimate and predict there to be 8.29.6 GW
of short-term load curtailment in California. This is signicantly
less than our estimates of the California residential air conditioner
resource potential ( E30 GW) for a number of reasons. First,
(Dyson et al., 2014) uses 1-h interval load data, which represents
average power consumption over one hour. This is in contrast to
our approach which models the hysteretic cycling behavior of
loads on timescales of seconds, therefore capturing much larger
load uctuations. Second, as shown in Table 5, PG&E has some of
the lowest resource potentials in the state and so extrapolating
PG&E estimates will underestimate the statewide potential. Third,
Dyson et al. (2014) attempted to include the effect of periods of
consumer absence, when outdoor air temperatures are high but
consumers are not home and so air conditioners are turned off. We
do not include this effect here.
Our estimated revenues given in Tables 6 and 7 vary signicantly based on type of TCL and type of ancillary service, with
regulation revenues always larger (and often much larger) than
spinning and non-spinning reserve revenues. Note that the capacities and revenues of air conditioners and heat pumps are highly
location-dependent. Electric water heaters are the most lucrative
TCL for spinning reserve ($9.70 per TCL per year) and heat pumps
are generally the most lucrative TCL for regulation ($2256 per TCL
per year). TCLs capable of acting as both an air conditioner and
heat pump could expect to earn revenue equal to the sum of their
estimated air conditioner and heat pump revenues ($2863 per
TCL per year for regulation), making participation in areas with
temperate climates more appealing.
In our previous work (Mathieu et al., 2012), we estimated
regulation and load following revenues using storage revenue
estimates from Eyer and Corey (2010). Their estimates are not

California-specic. Our regulation results were 23 times higher

than those in Tables 6 and 7. This is due to the fact that CAISO
ancillary service prices are generally lower than those in other
ISOs (MacDonald et al., 2012) and also due to Eyer & Corey's assumptions and methodology. Interestingly, in comparing our
previous results to the ones reported here, we found that while
both the air conditioner and heat pump revenue estimates decreased, the heat pump revenue estimates decreased more. This is
because in Mathieu et al. (2012) we used only an estimate of the
mean ancillary service price per MW, rather than time-varying
ancillary service price proles, to compute revenues, and air conditioner usage often coincides with periods of high ancillary service prices, while heat pump usage does not. Therefore, the results
reported here capture temporal effects more accurately. We do not
report load following revenue estimates in this paper because
California does not currently have a load following market. Instead, load following is accomplished by the real-time energy
market, which dispatches participants every 5 min. See (Mathieu
et al., 2014a) for estimates of energy cost savings achievable by TCL
aggregations arbitraging CAISO's real time energy market prices.
For residential TCLs to participate in load coordination programs without policy mechanisms to encourage their participation, the revenues must cover all costs and provide an incentive to
both the consumer and the load aggregator. Costs include communications and controls equipment that facilitates non-disruptive load coordination strategies and recurring costs including
operations and maintenance, incentive payments to the consumer,
energy required for sensing and communication, and prot for the
load aggregator. Recent research has focused on developing techniques that minimize the infrastructure required for load coordination. For example, Mathieu et al. (2013b) nd that load
performance in ancillary services markets is a function of the
amount and quality of information transmitted to and from each
load from an aggregator, and their lowest infrastructure case still
requires some investment in communication and control equipment that would enable TCLs to respond to coordination signals.
Before investing in non-disruptive load coordination, a load aggregator would need to determine if the revenues reported here
would be enough to cover all costs. However, the equipment required for participation in regulation and spinning/non-spinning

84

J.L. Mathieu et al. / Energy Policy 80 (2015) 7687

2014 Power Capacity

60
RF
EWH
HP
AC

GW

40
20
0

1000

2000

3000

4000

5000

6000

7000

8000

6000

7000

8000

2014 Energy Capacity

GWh

15
10
5
0

1000

2000

3000

4000
5000
hour of year

2020 Power Capacity

60
RF
EWH
HP
AC

GW

40
20
0

1000

2000

3000

4000

5000

6000

7000

8000

6000

7000

8000

2020 Energy Capacity

GWh

15
10
5
0

1000

2000

3000

4000
5000
hour of year

2020 Power Capacity Electrification Scenario

60
RF
EWH
HP
AC

GW

40
20
0

1000

2000

3000

4000

5000

6000

7000

8000

7000

8000

2020 Energy Capacity Electrification Scenario

GWh

15
10
5
0

1000

2000

3000

4000
5000
hour of year

6000

Fig. 5. Power and energy capacities of residential TCLs in California. (a) 2014, (b) 2020 and (c) 2020 electrication scenario.

Table 6
Mean regulation (reg) and spin/non-spin (sp/ns) power capacities (P ) of aggregations of 1000 heterogeneous refrigerators and electric water heaters, and
revenue estimate statistics (mean , standard deviation ). Revenues are reported
per TCL per year and given in USD 2014.

Refrigerators
Electric water
heaters

Preg

reg (reg )

P sp/ns

sp (sp )

ns (ns )

(kW)

($/TCL-yr)

(kW)

($/TCL-yr)

($/TCL-yr)

98
210

$11.00 (3.50)
$24.00 (7.80)

98
160

$4.30 (1.90) $0.53 (0.35)

$9.70 (4.30) $1.20 (0.80)

reserve would also be useful for other load control programs, such
as PG&E's SmartAC Program (PG&E, 2014), and so load aggregators/consumers may be able to leverage multiple revenue

streams to recoup equipment costs.

Consumer choices, policy changes, and new policy mechanisms
could increase the revenue potential of TCLs. For example, if a
consumer doubles the size of his/her TCL's temperature dead-band
he/she approximately doubles the TCL's energy capacity. Additionally, if consumers increase thermal mass inside their refrigerators and houses, they also can increase their energy capacity. In cases when the energy constraint is the binding constraint
in (10) or (11), increasing temperature dead-bands and thermal
mass increases regulation and spin/non-spin power capacities and
revenues. Additional policy changes and mechanisms include the
following.

The equipment required for TCL participation in ancillary service markets would also enable TCLs to offer other services to

J.L. Mathieu et al. / Energy Policy 80 (2015) 7687

Table 7
Mean regulation (reg) and spin/non-spin (sp/ns) power capacities (P ) of aggregations of 1000 heterogeneous air conditioners and heat pumps, and revenue
estimate statistics (mean , standard deviation ), by climate zone. Revenues are

85

thermostats have adjustable dead-bands would enable consumers to increase their thermostat's dead-band width, increasing their energy capacity and, subsequently, the statewide
resource potential.

reported per TCL per year and given in USD 2014.

Climate
zone

Preg

reg (reg )

P sp/ns

sp (sp )

ns (ns )

($/TCL-yr)

(kW)

($/TCL-yr)

($/TCL-yr)

Air conditioners
1
0
2
110
3
110
4
60
5
3.6
6
110
7
220
8
31
9
160
10
200
11 & 12
160
13
51

$0.00 (0.00)
$12.00 (4.00)
$12.00 (4.00)
$7.00 (2.00)
$0.33 (0.11)
$12.00 (4.00)
$32.00 (9.80)
$3.70 (1.30)
$22.00 (5.80)
$29.00 (8.70)
$22.00 (5.80)
$6.20 (1.90)

0
15
15
8.4
0.56
15
32
6.0
22
29
22
9.5

$0.00 (0.00)
$0.90 (0.32)
$0.90 (0.32)
$0.54 (0.16)
$0.02 (0.01)
$0.90 (0.32)
$2.40 (0.72)
$0.37 (0.14)
$1.70 (0.47)
$2.10 (0.64)
$1.70 (0.47)
$0.60 (0.21)

$0.00 (0.00)
$0.19 (0.17)
$0.19 (0.17)
$0.14 (0.14)
$0.01 (0.01)
$0.19 (0.17)
$0.39 (0.34)
$0.07 (0.08)
$0.35 (0.32)
$0.37 (0.34)
$0.35 (0.32)
$0.12 (0.12)

Heat pumps
1
2
3
4
5
6
7
8
9
10
11 & 12
13

$56.00 (19.00)
$28.00 (7.50)
$28.00 (7.50)
$32.00 (8.50)
$35.00 (11.00)
$28.00 (7.50)
$31.00 (8.60)
$30.00 (9.80)
$25.00 (7.90)
$34.00 (10.00)
$25.00 (7.90)
$22.00 (7.30)

76
42
42
47
49
42
37
34
30
40
30
26

$2.70 (1.30)
$1.30 (0.68)
$1.30 (0.68)
$1.50 (0.74)
$1.60 (0.95)
$1.30 (0.68)
$1.50 (0.63)
$1.50 (0.78)
$1.30 (0.62)
$1.70 (0.75)
$1.30 (0.62)
$1.10 (0.56)

$0.35 (0.23)
$0.13 (0.10)
$0.13 (0.10)
$0.16 (0.12)
$0.16 (0.12)
$0.13 (0.10)
$0.12 (0.09)
$0.11 (0.07)
$0.10 (0.07)
$0.13 (0.09)
$0.10 (0.07)
$0.08 (0.05)

(kW)

590
330
330
360
380
330
280
260
220
310
220
200

power systems, for example, distribution system support and

contingency reserves. Additionally, it would enable loads to
arbitrage real time electricity market prices (Mathieu et al.,
2013a, 2014a). Changing policy to allow load aggregators to
participate in these services and markets would give them
more potential revenue streams that they could leverage when
regulation and spinning/non-spinning reserve prices are low.
TCL aggregations respond to coordination signals much faster
than most power plants because they lack ramp rate constraints. Additionally, we expect them to have less environmental impact than the power plants that traditionally provide
reserves, which, when ramping, emit NOx and SOx and operate
in inefcient regimes (Katzenstein and Apt, 2009). Policies that
reward resources with superior performance or positive attributes could increase participation of TCL aggregations in ancillary service markets. The Federal Energy Regulatory Commission's Order 755 (FERC, 2011), which requires regulation
resources be paid based upon performance, is one step in this
direction.
The cost of retrotting TCLs with communications and controls
equipment is likely much higher than the incremental cost of
adding this equipment during manufacturing. Communications
and appliance standards2 would reduce infrastructure costs,
increasing the amount of revenue that could go towards recurring costs. A number of agencies and organizations are
currently developing smart grid interoperability standards including residential appliance standards. Communication standards should not only enable interoperability but also consider
privacy and security. Appliance standards could also increase
resource potentials. For example, standards requiring all

2
Note that by appliance standards we are not referring to minimum energy
performance standards but rather communication, interoperability, and controllability standards enabling load coordination.

An important caveat to our discussion is that, given the structure of today's U.S. electricity markets, increasing TCL participation
in ancillary services markets could lower the market clearing
price, decreasing potential revenues. In fact, Eyer and Corey (2010)
estimate the 10-year maximum regulation market potential in
California to be only 80 MW. Thus, our revenue potential analysis
is only valid for small participation rates.
In deriving our resource and revenue estimates, we have attempted to strike a balance between model accuracy and computational simplicity. As with all similar analyses, we had to make
a large number of assumptions. We have discussed the impact of
many of these assumptions in the paper, and believe that the results are not systematically biased high or low.

5. Conclusion and policy implications

We have developed methods to estimate the amount of energy
storage that could be provided by non-disruptive coordination of
residential TCLs. Specically, we modeled aggregations of TCLs as
virtual energy storage devices with time-varying energy and
power capacities. We have also developed methods to estimate the
revenue load could expect to earn by participating in ancillary
services markets. We used these methods along with publicly
available data to develop order-of-magnitude estimates for the
technical resource potential and ancillary service revenue potential of air conditioners, heat pumps, refrigerators, and electric
water heaters in California. We nd that the technical resource
potential is currently approximately 1040 GW/812 GWh (depending upon the time of year), which is signicantly more than
that required by California's energy storage mandate of 1325 MW
by 2020.
We nd that revenue potentials for TCL aggregations providing
regulation range from $0 to $56 per TCL per year depending upon
TCL type and, in the case of air conditioners and heat pumps, climate zone. Spinning reserve revenue potentials range from $0 to
$9.70 per TCL per year, while non-spinning reserve revenue potentials range from $0 to $1.20 per TCL per year. Refrigerators are a
promising resource because they have a large statewide resource
potential and are available throughout the year. In contrast, the
availability of air conditioners is a function of outdoor air temperature and therefore highly variable. Electric water heaters and
heat pumps have signicant resource and revenue potential, but,
currently, both exhibit low saturation rates across California. While
the revenue potential for heat pumps is generally higher than that
of electric water heaters, electric water heaters are available
throughout the year. On the other hand, heat pump availability is a
function of outdoor air temperature, and as with air conditioners
highly variable. TCLs capable of acting as both heat pumps and
air conditioners have the greatest revenue potential at $2863 per
TCL per year for regulation.
Consumers can increase their revenue potential by
(i) increasing the width of the their TCLs' temperature dead-bands
(i.e., increasing the range of allowable internal temperatures) and
(ii) increasing the thermal mass inside of their households and
refrigerators. Both of these actions increase a TCL's energy capacity, which, for an energy-constrained TCL, increases its regulation
and spin/non-spin power capacity.
The revenues from participation in ancillary services markets
would need to cover the upfront and recurring costs of non-disruptive load coordination and provide both sufcient prot for the

86

J.L. Mathieu et al. / Energy Policy 80 (2015) 7687

load aggregator and sufcient incentives for the consumers. There

are several possible policy changes/mechanisms that would increase revenue potentials. For example, removing barriers to participation in other power system services and markets would allow TCLs to leverage other value streams. Additionally, new policies like the Federal Energy Regulatory Commission's Order 755
(FERC, 2011), which compensates regulation service providers as a
function of performance, benets TCL aggregations, which are able
to respond quickly and accurately to coordination signals. Similar
policies for other ancillary services, and policies that compensate
service providers as a function of other positive attributes, such as
environmental impact, may also benet TCL aggregations. Lastly,
appliance standards enabling aggregator-to-load communication
would reduce the upfront costs associated with non-disruptive
load coordination, and standards requiring all thermostats to have
adjustable dead-band widths would allow consumers to determine their preferred amount of temperature exibility.

References
Alcorn, B., Ciminelli, M., Fugate, N., Gautam, A., Kavalec, C., Sullivan, K., WeingGutierrez, M., 2013. California Energy Demand 20142024: Revised Forecast,
vol. 1: Statewide Electricity Demand, End-User Natural Gas Demand, and Energy Efciency. Technical Report CEC-200-2013-004-SD-V1-REV, California
Energy Commission.
Alcorn, B., Ciminelli, M., Fugate, N., Gautam, A., Kavalec, C., Sullivan, K., WeingGutierrez, M., 2013. California Energy Demand 20142024: Revised Forecast,
vol. 2: Electricity Demand by Utility Planning area. Technical Report CEC-2002013-004-SD-V1-REV, California Energy Commission.
Bashash, S., Fathy, H., 2013. Modeling and control of aggregate air conditioning
loads for robust renewable power management. IEEE Trans. Control Syst.
Technol. 21 (4), 13181327.
CAISO, 2006. Spinning reserve and non-spinning reserve. California Independent System Operator. http://www.caiso.com/Documents/SpinningReserveandNonSpinnin
gReserve.pdf.
CAISO, 2009. Update on participating load functionality for markets and performance initiative. California Independent System Operator. http://www.caiso.
com/Documents/UpdateonParticipatingLoadFunctionality_Market
sandPerformanceInitiative.pdf.
CAISO, 2011. Regulation energy management draft nal proposal. California Independent System Operator. http://www.caiso.com/Documents/Revised
DraftFinalProposal-RegulationEnergyManagement-Jan13_2011.pdf.
CAISO, 2014. Non-generator resource ((NGR) and regulation energy management
(REM) overviewPhase 1. California Independent System Operator. http://
www.caiso.com/Documents/NGR-REMOverview.pdf.
CAISO, 2014. Open access same-time information system (OASIS). California Independent System Operator. http://oasis.caiso.com.
Callaway, D., 2009. Tapping the energy storage potential in electric loads to deliver
load following and regulation, with application to wind energy. Energy Convers. Manage. 50, 13891400.
Callaway, D., Hiskens, I., 2011. Achieving controllability of electric loads. Proc. IEEE
99 (1), 184199.
CCST, 2011. California's energy future: the view to 2050. California Council on Science and Technology. http://ccst.us/publications/2011/2011energy.pdf.
CEC, 2011. California building climate zone map. California Energy Commission.
http://www.energy.ca.gov/maps/renewable/building_climate_zones.html.
Chong, C.-Y., Debs, A., 1979. Statistical synthesis of power system functional load
models. In: IEEE Conference on Decision and Control. pp. 264269.
CPUC, 2008. Market redesign and technology upgrade (MRTU). California Public
Utilities Commission. http://www.cpuc.ca.gov/PUC/energy/wholesale/01a_ca
wholesale/MRTU/.
CPUC, 2013. Decision adopting energy storage procurement framework and design
program. California Public Utilities Commission. http://docs.cpuc.ca.gov/Pub
lishedDocs/Published/G000/M079/K533/79533378.PDF.
CPUC, 2013. RPS program overview. California Public Utilities Commission. http://
www.cpuc.ca.gov/PUC/energy/Renewables/overview.htm.
Dyson, M.E.H., Borgeson, S.D., Tabone, M.D., Callaway, D.S., 2014. Using smart meter
data to estimate demand response potential, with application to solar energy
integration. Energy Policy 74, 607619. http://dx.doi.org/10.1016/j.
enpol.2014.05.053.
EERE, 2012. EnergyPlus Energy Simulation Software: Weather Data. US Department
of Energy, Energy Efciency and Renewable Energy. http://apps1.eere.energy.
gov/buildings/energyplus/cfm/weather_data.cfm.
EERE, 2014. Estimating Appliance and Home Energy Usage. US Department of Energy, Energy Efciency and Renewable Energy. http://www.energysavers.gov/
your_home/appliances/index.cfm/mytopic 10040.
EIA, 2005. Residential Energy Consumption Survey (RECS): 2009 RECS Survey Data.
US Energy Information Agency. http://www.eia.gov/consumption/residential/

data/2005/.
EIA, 2009. Residential Energy Consumption Survey (RECS): 2009 RECS Survey Data.
US Energy Information Agency. http://www.eia.gov/consumption/residential/
data/2009/.
Eyer, J., Corey, G., 2010. Energy Storage for the Electricity Grid: Benets and Market
Potential assessment Guide. Technical Report SAND2010-0815, Sandia National
Laboratory, SAND2010-0815.
FERC, 2011. Order no. 755, Frequency Regulation Competition in Organized
Wholesale Power Markets. Federal Energy Regulatory Commission. http://
www.ferc.gov/whats-new/comm-meet/2011/102011/E-28.pdf.
Halamay, D., Brekken, T., Simmons, A., McArthur, S., 2011. Reserve requirement
impacts of large-scale integration of wind, solar, and ocean wave power generation. IEEE Trans. Sustain. Energy 2 (July (3)), 321328.
Ihara, S., Schweppe, F., 1981. Physically based modeling of cold load pickup. IEEE
Trans. Power Apparatus Syst. PAS 100 (9), 41424150.
Katzenstein, W., Apt, J., 2009. Air emissions due to wind and solar power. Environ.
Sci. Technol. 43, 253258.
Kavalec, C., Gorin, T., 2009. California Energy Demand 20102020: Staff Revised
Forecast. Technical Report CEC-200-2007-015-SF2, California Energy
Commission.
KEMA-XENERGY, June 2004. California Statewide Residential Appliance Saturation
Survey. Technical Report CEC400-04-009, California Energy Commission.
Kundu, S., Sinitsyn, N., Backhaus, S., Hiskens, I., August 2011. Modeling and control
of thermostatically controlled loads. In: Proceedings of 17th Power Systems
Computation Conference. Stockholm, Sweden.
Loutan, C., Hawkins, D., Blatchford, J., de Mello, P., DeShazo, G., Dukes, D., Fuller, D.,
Jetmalani, V.N., Le, D., Stone, S., Wong, J., Zhang, Y., Zhu, S., Clark, K., D'Aquila, R.,
Miller, N., Ma, J., Makarov, Y., Brower, M., 2007. Integration of Renewable Resources: Transmission and Operating Issues and Recommendations for Integrating Renewable Resources on the California Iso-Controlled Grid. Technical
Report, California Independent System Operator Renewables Workgroup.
Lutz, J., Ghatikar, G., Edelson, E., Meyers, S., October 2006. Hot Water Draw Patterns:
Findings from Field Studies. Technical Report Appendix H of Water Heaters
and Hot Water Distribution Systems CEC-500-2005-007-APA, Lawrence Berkeley National Laboratory. http://www.energy.ca.gov/2008publications/CEC500-2008-082/CEC-500-2008-082-APA.PDF.
MacDonald, J., Cappers, P., Callaway, D., Kiliccote, S., 2012. Demand response providing ancillary services: a comparison of opportunities and challenges in the
US wholesale markets. In: Proceedings of the Grid Interop (LBNL-5958E). Irving,
TX.
Makarov, Y., Loutan, C., Ma, J., De Mello, P., 2009. Operational impacts of wind
generation on California power systems. IEEE Trans. Power Syst. 24 (2),
10391050.
Marshall, L., Gorin, T., 2007. California Energy Demand 20082018: Staff Revised
Forecast. Technical Report, California Energy Commission, CEC-200-2007-015SF2.
Mathieu, J., Dyson, M., Callaway, D., 2012. Using residential electric loads for fast
demand response: the potential resource and revenues, the costs, and policy
recommendations. In: Proceedings of the ACEEE Summer Study on Energy Efciency in Buildings, vol. 1. Pacic Grove, CA, pp. 189203.
Mathieu, J., Kamgarpour, M., Lygeros, J., Andersson, G., Callaway, D., 2014. Arbitraging intraday wholesale energy market prices with aggregations of thermostatic loads. IEEE Trans. Power Syst, (in press). http://dx.doi.org/10.1109/TPWRS.
2014.2335158.
Mathieu, J., Kamgarpour, M., Lygeros, J., Callaway, D., 2013. Energy arbitrage with
thermostatically controlled loads. In: Proceedings of the European Control
Conference. Zrich, Switzerland, pp. 25192526.
Mathieu, J., Koch, S., Callaway, D., 2013b. State estimation and control of electric
loads to manage real-time energy imbalance. IEEE Trans. Power Syst. 28 (1),
430440.
Mathieu, J., Rasmussen, T., Srensen, M., Jhannsson, H., Andersson, G., 2014.
Technical resource potential of non-disruptive residential demand response in
denmark. In: Proceedings of the IEEE Power & Energy Society General Meeting.
National Harbor, MD.
Mathieu, J.L., 2012. Modeling, Analysis, and Control of Demand Response Resources
(Ph.D. thesis), University of California, Berkeley.
Morgan, M.G., Small, M., 1992. Uncertainty: A Guide to Dealing with Uncertainty in
Quantitative Risk and Policy Analysis. Cambridge University Press, New York,
NY.
Palmgren, C., Stevens, N., Goldberg, M., Bames, R., Rothkin, K., 2010. 2009 California
Residential Appliance Saturation Survey. Technical Report, Prepared for the
California Energy Commission by KEMA, CEC-200-2010-004.
Perfumo, C., Kofman, E., Braslavsky, J., Ward, J., 2012. Load management: modelbased control of aggregate power for populations of thermostatically controlled
loads. Energy Convers. Manag. 55, 3648.
PG&E, 2014. SmartAC program. Pacic Gas and Electric Company. http://www.pge.
com/en/myhome/saveenergymoney/plans/smartac/index.page.
Schneider, K., Fuller, J., Chassin, D., 2011. Multi-state load models for distribution
system analysis. IEEE Trans. Power Syst. 26 (4), 24252433.
State of California, 2010. Public Utilities Code Section 28352839:
Chapter 7.7 Energy Storage Systems. State of California Legislative Information.
http://leginfo.legislature.ca.gov/faces/codes_displayText.xhtml?law
Code PUC&division 1.&title &part 2.&chapter 7.7.&article .
Sweet, B., 2013. California's rst-in-nation energy storage mandate. IEEE Spectrum.
http://spectrum.ieee.org/energywise/energy/renewables/californias-rst
innation-energy-storage-mandate.

J.L. Mathieu et al. / Energy Policy 80 (2015) 7687

Totu, L.C., Leth, J., Wisniewski, R., 2013. Control for large scale demand response of
thermostatic loads. In: American Control Conference. pp. 50235028.
Vrettos, E., Koch, S., Andersson, G., 2012. Load frequency control by aggregations of
thermally stratied electric water heaters. In: IEEE PES Innovative Smart Grid
Technologies (ISGT) Europe. Berlin, Germany.
WECC, 2007. AC Unit Model Specications. Technical Report, Western Electricity
Coordinating Council Load Modeling Task Force. http://www.wecc.biz/com

87

mittees/StandingCommittees/PCC/TSS/MVWG/032708/Lists/Minutes/1/AC%
20Unit%20Model%20Specications.pdf.
Zhang, W., Lian, J., Chang, C., Kalsi, K., 2013. Aggregated modeling and control of air
conditioning loads for demand response. IEEE Trans. Power Syst. 28 (4),
46554664.