You are on page 1of 9

IEEE TRANSACTIONS ON SMART GRID, VOL. 2, NO.

2, JUNE 2011 411

Appliance Commitment for Household


Load Scheduling
Pengwei Du, Member, IEEE, and Ning Lu, Senior Member, IEEE

Abstract—This paper presents a novel appliance commitment Actual market price at time interval
algorithm that schedules thermostatically controlled household
($/MWh).
loads based on price and consumption forecasts considering
users’ comfort settings to meet an optimization objective such as Forecasted market price at time interval
minimum payment or maximum comfort. The formulation of an ($/MWh).
appliance commitment problem is described using an electrical
water heater load as an example. The thermal dynamics of heating Thermal energy needed to heat up the water
and coasting of the water heater load is modeled by physical heater (J).
models; random hot water consumption is modeled with statistical
methods. The models are used to predict the appliance operation Time at the th step (hour).
over the scheduling time horizon. User comfort is transformed to a
set of linear constraints. Then, a novel linear-sequential-optimiza- Status of water heater over period
tion-enhanced, multiloop algorithm is used to solve the appliance (1—on, 0—off).
commitment problem. The simulation results demonstrate that the
algorithm is fast, robust, and flexible. The algorithm can be used Sorted (monotonic) price curve ($/MWh).
in home/building energy-management systems to help household Lower and upper limit of hot water temperature
owners or building managers to automatically create optimal load
operation schedules based on different cost and comfort settings
setting F .
and compare cost/benefits among schedules. Constants:
Index Terms—Appliance commitment, end-user comfort, home Amount of C by which the temperature of the
energy management, load scheduling.
added water should be increased.
Specific heat of water J/kg C .
NOMENCLATURE
Variables: Time elapsed per step (i.e., )
(hour).
Ambient temperature/cold water inlet
temperature F Water heater thermal capacitance C .

Temperature of hot water at time, F . Mass of water in the full tank (gallons).
Power rating of water heater (kW).
Temperature of hot water at time F .
Heat loss of the electric water heater (J).
Current temperature of hot water (after water
draw) F . Electric water heater capacity (kW).
Electricity payment ($). Water heater thermal resistance C/kW
Price threshold used to determine control logic Mass of the added water at time (kg) (1
for water heater ($/MWh). gallon water 3.785 kg).
Temperature threshold used to determine
control logic for water heater F . I. INTRODUCTION

Duration the water heater needs to be turned on


(minutes).
Demand of hot water drawn (or mass of water
T RADITIONALLY, a power system is only scheduled for
operation on generation resources because the majority
of power system loads are neither controllable nor measurable
from cold water inlet) at time (gallons). at the required time resolution. In addition, the nondifferential
retail electricity price provides little incentive for consumers to
schedule their energy consumption.
Manuscript received November 02, 2010; revised February 22, 2011 and
From 2010 to 2012, millions of smart meters, sensors, and
March 19, 2011; accepted March 24, 2011. Date of publication May 10, 2011;
date of current version May 25, 2011. Paper no. TSG-00203-2010. automatic control devices will be deployed in electric power
The authors are with Pacific Northwest National Laboratory, Richland, WA distribution grids down to residential and commercial build-
99352 USA (e-mail: pengwei.du@pnl.gov; ning.lu@pnl.gov).
ings via two-way communication networks. Retail pricing
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. schemes, such as real-time pricing (RTP), time-of-use (TOU),
Digital Object Identifier 10.1109/TSG.2011.2140344 and critical peak pricing (CPP), are being designed by utilities

1949-3053/$26.00 © 2011 IEEE


412 IEEE TRANSACTIONS ON SMART GRID, VOL. 2, NO. 2, JUNE 2011

to provide incentives for consumers to actively participate in The formulation of an appliance commitment problem is de-
demand response (DR) programs. Furthermore, it is possible scribed in this paper using the thermostatically controlled elec-
to control and monitor the on/off, cycling, or mode switching tric water heater (EWH) load as an example.
of appliances wirelessly with a home energy management Although the scheduling of an EWH has been studied ex-
system by developing smart appliances and deploying home tensively ([3]–[9]), and there is a rich literature on how to
area networks (HANs). For example, the thermostat settings of use the tariff structure to minimize energy payment under
a heating, ventilating, and air conditioning (HVAC) system can different rate structures, the customer comfort level has been
be programmed remotely via a ZigBee network; a smart dryer rarely discussed. Furthermore, most of the existing scheduling
can be switched from the “normal” mode to “energy saving” problems were solved by complex nonlinear optimization
modes to reduce its energy consumption. methods, such as Lagrange methods. A linear approach was
The smart grid technologies are enablers to schedule loads at proposed by Schweppe et al. [3] and Daryanian et al. [4]. An
the consumer level to save energy, reduce cost, and help grid op- optimization framework was developed by Mohsenian-Rad
eration; however, it is impractical to request a consumer, who is and Leon-Garcia [8] to minimize the household’s electricity
neither an economist nor an experienced grid operator, to create payment by optimally scheduling the operation and energy
an optimal schedule from the many possibilities. Therefore, an consumption for each appliance, while considering the waiting
automatic load scheduling method that requires little consumer time as a comfort setting for the operation of each appliance.
attention to set up and maintain and allows consumers to com- A distributed energy resources scheduling algorithm was intro-
pare costs/benefits with different load schedules, is needed to duced by Pedrasa et al. [9] to maximize the owner’s revenue by
gain consumer acceptance on consumer-level DR programs. It considering the deviation from the user-desired temperature as
is also needed to motivate consumers’ active participation in an undelivered energy service. Those approaches have several
shaping their energy consumption for the benefit of themselves limitations: 1) in Schweppe et al. [3] and Daryanian et al. [4],
as well as for the efficient, reliable, and economical operation a linear thermal model for the water heater was used; 2) no
of the power grid. mechanism was used to tackle the prediction errors in price
The main contribution of this paper is the development of a and hot water usage; and 3) they did not explicitly consider the
novel appliance commitment algorithm that schedules thermo- user choice on thermostat settings as a constraint in terms of a
statically controlled appliances (TCAs) based on price and con- time-varying temperature band.
sumption forecasts. The appliance commitment approach specifies a time-varying
The appliance commitment, similar to the unit commitment temperature range to reflect consumer choices on the appli-
problem [1], finds an optimal schedule for each device (gen- ances’ thermostat settings and their perception of comfort
erating unit for unit commitment and appliances for appliance constraints. In addition, the two-step optimization framework
commitment) over a given period of time based on a number proposed in this paper can adjust the appliance operation
of operational constraints and economical considerations. The schedule to handle uncertainties that are caused by forecast
formulation of an appliance commitment problem differs from errors in energy prices and hot water consumption.
that of the unit commitment in two aspects: 1) the electricity The thermal dynamics of heating and coasting of the EWH
consumption of an appliance is partially driven by random con- load is simulated using physical models; random hot water con-
sumer usage (e.g., random hot water consumption), and appli- sumption is modeled with statistical methods. The models are
ances have fewer power output levels, and 2) the formulation of used to predict appliance operation over the scheduling time
the TCA constraints is generally more difficult than the formula- horizon. User comfort is transformed to a set of linear con-
tion of generator constraints (for example, user comfort settings straints. The appliance commitment is solved with a multiple-
need to be reflected as a set of linear or nonlinear constraints in looping algorithm introduced by Lu et al. [10] and enhanced
the optimization problem). Although the algorithm was devel- by a novel linear sequential optimization process, which is easy
oped to schedule TCAs, with a little modification, the algorithm to implement, robust to variations of the forecasted prices, and
can easily be extended to schedule other types of loads, as well flexible to handle comfort constraints.
as distributed generators or energy storage devices. This paper is organized as follows. The formulation of the
Agent-based approaches have been applied to design and op- appliance commitment problem, including the thermal dynamic
erate individual smart appliances [2]. One example of the ap- model of the water heater load, the statistic model of hot water
proach is to let an appliance respond to the deviation of power consumption, and the introduction of comfort constraints, is
grid frequency or bid into the market to negotiate the price with described in Section II. The solution process of the appliance
others. Compared with the appliance commitment approach, the commitment is illustrated in Section III. The simulation results
agent-based approach has the following limitations: 1) the in- are discussed in Section IV. The conclusions are provided in
telligence of each agent on an appliance is very limited, and 2) Section V.
the coordination among appliances is very difficult. Most im-
portantly, because the appliance commitment algorithm has a II. THE FORMULATION OF THE APPLIANCE
scheduling and load coordination process based on price and COMMITMENT PROBLEM
energy consumption forecasts over a planning horizon, (e.g., 24 Household appliances can be classified into three groups:
h), it outperforms the agent-based approach by being more flex- controllable thermostatically controlled appliances (C-TCAs),
ible; also, it results in more energy savings. controllable nonthermostatically controlled appliances
DU AND LU: APPLIANCE COMMITMENT FOR HOUSEHOLD LOAD SCHEDULING 413

TABLE I
TYPICAL VALUES FOR A WATER HEATER

A typical water heater comes in standard tank sizes that may vary from
30 gal (113.58 L) to 80 gal (302.88 L), the most common being 50 gal
(189.25 L) [7].

the period (i.e., ), its temperature at the time,


, increases to , given by
Fig. 1. A typical thermal characteristic curve of an EWH load.

(1)
(non-TCAs), and noncontrollable appliances. Noncontrol- Otherwise, when an EWH is off over the period
lable appliances cannot be scheduled; therefore, they are (i.e., ), the temperature at the time, , drops because
modeled by a load profile that is forecast with historical data. of power loss and becomes
Controllable non-TCAs (such as washers or dryers) are nor-
mally discontinuous loads that are straightforward to schedule (2)
in an appliance commitment problem. The C-TCA loads, such
as HVACs and water heaters, have thermal storage capabilities Most heaters run because of the household’s usage of hot
and are, traditionally, good candidates for consumer-level DR water rather than heater loss. When a household uses hot water,
programs. The scheduling of the C-TCA loads has to account cold water refills at the bottom of the tank. The temperature of
for thermal dynamics, random consumption, and consumer the water after adding inlet water is given by
comfort. Therefore, the capability of scheduling C-TCA loads
is at the core of the appliance commitment algorithm. To (3)
simplify our presentation, the scheduling of an EWH load is
used as an example to illustrate the formulation of an appliance Together, (1), (2), and (3) describe the thermal dynamics of
commitment problem. the EWH, which can be rewritten as a function of thermal pa-
To develop an optimal schedule over a time horizon, the ap- rameters, hot water demand, ambient temperature, and on/off
pliance commitment requires a price forecast, a range of thermo- status
stat settings, and the characteristics of the electric water heater.
In addition, the demand for hot water needs to be predicted (4)
from time-based heating data. In this paper, the primary objec-
tive of appliance commitment for a water heater is to minimize
the electricity bill over the next 24 h subject to constraints on B. The Parameter Estimation of the EWH Thermal Model
comfort level and meeting the predicted hot water demand. The The EWH thermal model is characterized by thermal resis-
comfort zone is defined by the comfort constraints, which re- tance and thermal capacitance. These thermal coefficients can
flect the range of the hot water temperature set by consumers. be estimated with statistical and regression techniques by fit-
An increase in the comfort zone will provide more flexibility in ting the observed performance data, such as temperature and
scheduling, resulting in higher savings. This comfort zone can water demands, to (1)–(4). A detailed discussion on how to use
be time-varying. the maximum likelihood method to estimate system parameters
can be found in Pahwa and Brice [18].
A. The Thermal Dynamic Modes of the EWH An alternative way is to choose the thermal coefficients for
water heaters from the 2008 ASHRAE Handbook [19]. For ex-
The modeling of an EWH operation requires a thermal dy- ample, typical values for a residential water heater are shown in
namic model that describes its heat exchange with the environ- Table I.
ment and with cold water inflows [11]–[17]. The thermal model
we used is based on an equivalent thermal parameter (ETP) ap- C. The Random Hot-Water-Consumption Model
proach [13]–[17]. Fig. 1 shows the water temperature inside
the EWH tank over time. The rising curves indicate that the Hot water consumption varies from one household to another,
EWH is “on,” and the falling curves represent the standby (or depending on the number of users, water-consumption habits
cooling-down) periods when the heater is “off.” As the EWH (the frequency of hot water use and the duration of the water
cycles, the water temperature in the tank rises and falls accord- consumption), and the facilities of a house (bath, shower, etc.).
ingly. The upper and lower limits represent the dead-band of the A typical hourly hot water usage in summer and winter is shown
thermostat around the thermostat set-point. in Fig. 2.1
The thermal dynamic behavior of an EWH can be described 1Real measurement data collected in the Pacific Northwest GridWise Testbed
by differential equations [13]–[15]. When the EWH is on over project [20] were used in our study.
414 IEEE TRANSACTIONS ON SMART GRID, VOL. 2, NO. 2, JUNE 2011

III. SOLVING THE APPLIANCE COMMITMENT PROBLEM


This section illustrates the process for solving the appliance
commitment problem.

A. Problem Formulation
Assume that the scheduling of an EWH load is over a 24-h
period. The control objective is to minimize the energy bill sub-
ject to comfort constraints that reflect a hot water temperature
band set by consumers. The hot water consumptions during the
scheduling period need to be considered as well. The decision
Fig. 2. Typical hot water usages of a winter/summer day (blue: winter; variables are the status of hot water over the next 24 h.
red: summer). Mathematically, the appliance commitment can be formu-
lated as
The historical data of hot water demand can be collected to
predict future hot water usage. A flow meter can provide his- (5a)
torical data on the time and the volume of water drawn from
the water heater. Alternatively, the hourly electricity consump-
tion for an EWH can provide information on the periods during subject to the following constraints:
which it is used. Thus, the hot water demand value can be calcu-
(5b)
lated because the electricity usage of a water heater is at a con-
stant rate when it is on. Future demands can be predicted from (5c)
the evaluated demand patterns of previous days. Section IV has
The objective in (5a) is to minimize the cost for energy con-
a detailed discussion on EWH load profile forecasting.
sumption over the next 24 h. The constraints in (5b) and (5c)
are the hot water thermal dynamics and the temperature bounds
specified by users to reflect their comfort. As the temperature
D. The Comfort Constraints bounds become more restrictive, there is less freedom for the
optimal solution. Therefore, the energy consumption scheduling
Customers may adopt strategies for their water heaters to re- problem is formulated as a nonlinear optimization problem that
duce their energy bills with dynamic pricing. Thus, in contrast aims to minimize the electricity payment subject to the user-
to the normal maintenance of the full-storage water heater, one comfort constraint.
can choose low-cost times for heating water and take advan- B. The Two-Step Scheduling Process
tage of storage capability to have enough hot water to meet the
demand at high-cost times. However, it might be necessary to The challenges of the load scheduling are as follows: 1) the
inconvenience the resident to reduce the energy bill. This incon- model of hot water thermal dynamics is nonlinear; 2) the price
venience can be that the water temperature is too low at the time forecasts may not be accurate; and 3) the exogenous demand for
of demand, or there is a long waiting time before the temperature hot water is stochastic. Therefore, the algorithm has to be flex-
is satisfactory. This conflict in two objectives, i.e., minimizing ible and robust to account for the aforementioned nonlinearity
cost and minimizing the negative impact over comfort, has been and uncertainties.
barely discussed in the literature. Addressing this requires a de- The model uncertainties can be explicitly handled by ap-
tailed thermal model of water heaters and an optimal framework plying stochastic dynamic programming [25]. However, this
that takes into account the operational constraints specified as method may be computationally time-consuming. Further-
the comfort level. more, the temperature bounds can be seriously violated for
Different residences have different preferable temperature unexpected variations in price or water demand. The above
ranges for their water heaters. The thermostat settings of water modeling issues motivate us to develop a two-step adjustment
heaters typically vary from 120 F to 160 F. Customers can process—day-ahead scheduling and real-time adjustment—as
adjust the setting to reflect the compromise between the com- shown in Fig. 3. First, use the day-ahead prediction of spot
fort level and the energy cost. There is also a safety concern prices and hot water usage to solve the deterministic optimal
when the temperature of the water heater is too high.2 In this problem over the next 24 h and then adjust the scheduling in
paper, the acceptable bounds on the temperature, which will real-time based on updated information.
be specified by the customer to reflect individual needs, are 1) Day-Ahead Scheduling: There is a similarity between op-
used as the operational constraint in the scheduling process. erating an EWH and an energy storage device. The preheating
Moreover, the result of the schedule as a function of different period of an EWH is similar to the pumping mode of a pumped-
comfort levels is calculated to offer flexibility to users. storage power station, and the coasting period is similar to its
generating mode. Therefore, optimal bidding strategies devel-
2Significant scalding risks exist at tap discharge temperatures greater than
oped in Lu et al. [10] for a pumped-storage hydro-turbine can
120 F. A few states require set-points in the 120–125 F range. For example,
§704.06 of the Wisconsin State Code requires a residential landlord to set the be used for applying the preheating and coasting control over
water heater temperature no higher than 125 F [7]. an EWH.
DU AND LU: APPLIANCE COMMITMENT FOR HOUSEHOLD LOAD SCHEDULING 415

As such, when the price is lower than , the heater


is “on”; otherwise, it is “off.”
The control logic in (9), , is
a solution to the relaxing deterministic optimal
problem without considering the comfort constraint,
which is given by

(10a)

Fig. 3. An illustration of the two-step scheduling process.


subject to

To minimize the payment, the preheating period should be (10b)


done during the least-price hours, and the coasting period should
cover the peak-price period. To maximize the comfort level, the Step 4: Check the solution with the status of the EWH
EWH should be switched on immediately to restore the temper- determined by (9). Stop if there is no violation of
ature after a significant portion of hot water is drawn from the the comfort band. Thus, the total payment, , is
tank, and cold water is refilled. Any delay will lead to a pro-
longed uncomfortable period or the violation of the tempera- (11)
ture-band constraint.
Therefore, the optimization process is as follows:
Otherwise, subdivide the time interval into
and where is the time when the EWH
Step 1: Obtain a forecasted, day-ahead, market clearing crosses its upper or lower comfort band.
price (MCP) curve.3 Step 5: After , set the temperature set-point of the water
Step 2: Based on the forecasted usage of hot water, estimate heater to saturated5 (either as the upper comfort
how long the EWH needs to be “on” in a time band or as the lower comfort band). For example,
interval to heat up the water in the tank. at midnight when the price is low, the EWH will
The amount of energy needed to restore the energy be switched on until its temperature reaches the
used is given in [10] as4 upper bound, i.e., a full tank of hot water is ready
for use. After that, the EWH remains off until
(6) the temperature of the hot water falls back into
the prespecified bound at the time, (i.e., next
The duration the water heater needs to be turned significant hot water draw).
on to provide is Step 6: Set and go back to Step 2.

min (7) 2) Real-Time Adjustment: The real-time market prices


usually deviate from the day-ahead forecast prices, and the
hot water demand may change in real-time. The commitment
This establishes a link between the required
of EWH needs to be adjusted to reflect the real-time changes.
operation time of an EWH and the hot water usage
Based on the updated hot water usages, the algorithm updates
of a household.
the amount of energy, , using (6) and then calculates the
Step 3: Find the price threshold, , based on how long the
new operating period . With the updated historical price
EWH needs to be operated.
information, resort the price curve to yield a new price curve,
The estimated price obtained in Step 1 in a time
. Thus, the updated price threshold, , is
interval can be sorted ascending to yield a
monotonic price curve, . Using (7), the period for (12)
the water heater to be on, , can be calculated.
Then, the price threshold, , will be determined
from the sorted price curve, , as
IV. SIMULATION RESULTS
(8) The simulation results are presented in this section. The pa-
rameters of the EWH model are given in Table I. The appliance
The control logic for the EWH can be described by commitment schedule starts from 12 a.m. and runs for 24 h to
12 a.m. the next day.
(9)
5When the EWH attemps to maintain its temperature near the upper or lower

3The
temperature limits, the high-frequency switching can occur. To overcome this,
market price is forecasted on an hourly basis over the next 24 h. a hyperthesis needs to be used ( F in this paper). However, depending
4The standby heat loss of commercially available heaters is very small [7]. on the system requirement, a large value may be needed like [7] F .
416 IEEE TRANSACTIONS ON SMART GRID, VOL. 2, NO. 2, JUNE 2011

Fig. 6. The hot water temperature profiles (Scheme A: no-control [red] and
Fig. 4. The RTP prices (blue: actual; red: day-ahead forecast). Scheme B: transactive control [blue]) (the upper and lower dot lines represent
the temperature limits for Scheme A).

C. Transactive Control Strategy


Assume that the upper temperature limit of an EWH is set at
160 F and the lower temperature limit at 140 F. If the ther-
mostat of the EWH is not price-sensitive (Scheme A), the water
temperature profile [Fig. 6 (red)] is simulated using the fore-
casted hot water usage (the red line in Fig. 5). The electricity
payment is $1.018 on this typical day based on the forecasted
Fig. 5. Forecast hot water usage (blue: actual; red: forecast). RTP prices (the red line in Fig. 4).
The performance of appliance commitment and the nonprice-
A. Day-Ahead Energy Price Forecast sensitive scheme is also compared to that of the transactive con-
In the simulation, the actual price signal used, , was col- trol strategy applied in the Pacific Northwest GridWise Testbed
lected in the Pacific Northwest GridWise Testbed project [20]. project. A transactive control strategy (Scheme B) applied to a
Normally, a day-ahead price forecast curve can be obtained thermostatically controlled HVAC was introduced [2]. The al-
from historical price data using different forecasting methods gorithm uses the current indoor temperature to calculate the bid
[21]–[24], [26]. price. After receiving the market clearing price, the adjusted
In our example, the forecasted price signal, , is obtained by zone set-point is calculated to reset the thermostat. The con-
adding a white noise, , (which follows a normal distribution) troller is given a range of daily price variations to adjust the
to the actual price signal, as in (13) thermostat set-point so that the thermostat reaches its minimum
setting when the price is at maximum. If the transactive control
(13) scheme is applied to the EWH with the forecast hot water con-
sumption and RTP prices, the electricity payment is $0.912 on
A daily mean error of 10% is assumed in this paper, which is this typical day. The hot water temperature profile is shown in
consistent with the forecasting accuracy reported by Contreras Fig. 6 (blue).
et al. [21], Hong and Hsiao [22], Nogales et al. [23], and Bohn The advantage of the transactive control is that it is easy to
[24], [26]. The RTP price forecast (red) and the actual price implement at the appliance-level. The disadvantage of the trans-
signal (blue) on a typical day are shown in Fig. 4. active control strategy is that it is a real-time control strategy for
individual appliances and cannot optimize the operation of mul-
B. A State Queueing Model of Water Heater tiple appliances over a time period.
An SQ model proposed by Lu et al. describes a water heater
behavior by tracking the state of the water heater over time D. Appliance Commitment
(see detailed discussions on this mode in Lu and Chassin [13] In this subsection, the two-step scheduling process introduced
and Lu et al. [14]). The evolution of the state of the water in Section III is applied on an EWH load to determine the day-
heater is mainly driven by the consumer behavior, which is the ahead schedule and perform real-time adjustments based on the
most important factor that causes the heater to turn on and off forecasted and actual RTP price and hot water usage, as shown
in a day. The customer consumptions are classified into two in Figs. 4 and 5.
categories: major consumptions and minor consumptions. A 1) Day-Ahead Schedule: Sort the forecasted RTP prices in
major consumption includes behaviors such as washing dishes ascending order to obtain a composite price curve, , as shown
or taking showers, which usually last for more than 5 min. in Fig. 7. The price threshold, , is calculated to be 12$/MWh,
A minor consumption includes behaviors such as washing and the total heater “on” time, , is 9.18 h to meet the heat
hands and washing fruit, which usually last for 1 or 2 min. losses and hot water demand. If comfort constraints are not con-
The probabilities of major and minor hot water consumptions sidered, the operational strategy for the EWH is very simple.
can be calibrated to minimize the difference between historical When the spot price is less than , the EWH is “on”; other-
demand data and simulation data. Therefore, from these proba- wise, it is “off.” The electricity payment is $0.248, significantly
bilities, simulations can be performed to yield the prediction of lower than that of the no-control case (Scheme A) or the trans-
the water usage (red) for the next day, as shown in Fig. 5. active-control case (Scheme B). However, as shown in Fig. 8,
DU AND LU: APPLIANCE COMMITMENT FOR HOUSEHOLD LOAD SCHEDULING 417

TABLE II
COMPARISON OF PERFORMANCE OF DIFFERENT CONTROL SCHEME

Fig. 7. The composite price curve (price threshold, ) (blue: day-ahead


forecast price; red: sorted day-ahead forecast price).

Scheme A: conventional thermostat; Scheme B: dynamic price-sensitive


thermostat; Scheme C: optimal schedule without considering users’
comfort; Scheme D: optimal schedule considering users’ comfort (uniform
temperature constraint); Scheme E: optimal schedule considering users’
comfort (time-varying temperature constraint)

In the second case, the temperature bands change with respect


to the time of day
Fig. 8. Scheme C: Hot water temperature as a result of the optimal minimum-
cost schedule without considering users’ comfort (the electricity payment is
$0.248). (15)

The hot water temperature (see Fig. 9(b)) decreases in late after-
noon as a result of relaxed constraints in the water temperature.
The electricity payment is $0.798. Compared with $0.811 in the
first case, the economical gain is marginal.
The performance of different control schemes is compared in
Table II. Compared with the no-control case (Scheme A), the
new schemes (Scheme D and E) can achieve over 20% in sav-
ings without compromising users’ comfort constraints. More-
over, this flexible framework can accommodate time-varying
Fig. 9. Hot water temperature F as a result of optimal schedule considering comfort settings.
users’ comfort. (a) Scheme D: uniform temperature constraint (the electricity 2) Real-Time Adjustment: The actual price is usually
payment is $0.811). (b) Scheme E: time-varying temperature constraint (the
electricity payment is 0.798). different from the forecast price (see Fig. 4). A comparison
between the sorted forecast prices and sorted actual prices is
the hot water temperature can be higher than 200 F or lower shown in Fig. 10. If the control logic that is determined based
than 100 F, which are unacceptable to consumers. Therefore, on a day-ahead forecast is used for the actual price signal,
the comfort constraints reflecting the consumer choices of the it results in about an additional 5 min heating time, and the
hot water temperature band must be included. actual electricity payment is $0.778. The payment difference
In the first case, assume that the temperature band is uniform is small; however, the hot water temperature deviations can be
over the whole scheduling period (24 h) very large because of forecasting errors in price and hot water
demand. As shown in Fig. 11, the water temperatures (the
when (14) red line) have crossed the upper and lower temperature limit
(140 F–160 F .
After solving the unconstrained case, check the first violation. A real-time adjustment is conducted as described in
As shown in Fig. 8, the violation of the upper temperature limit Section III-B. After the adjustment, the temperature pro-
occurs at h; therefore, the whole optimization time file (the blue line in Fig. 11) is significantly improved; the
period [0, 24] is separated into two: [0, 1.466] and [1.466, 24]. trade-off is that the cost is slightly higher ($0.831). Note that
Starting from h, the heater turns off and maintains around 9 a.m., the water temperature dropped below 140 F
its temperature until 4.916 h. Then, another unconstrained op- even though the water heater was switched on. This was caused
timization is carried out over a time period [4.916, 24]; after by a large hot-water consumption that occurred in that period.
it is done, the upper and lower temperature limits are checked As demonstrated, the two-step optimization approach can ef-
again. The procedure repeats until no violation of the tempera- fectively adjust the load schedule to reflect the price changes,
ture limits occurs. The resultant electricity payment is $0.811, thereby accounting for the forecasting error in the price in the
and the hot water temperature profile is shown in Fig. 9(a). problem formulation.
418 IEEE TRANSACTIONS ON SMART GRID, VOL. 2, NO. 2, JUNE 2011

by modeling the appliance thermodynamic process and ex-


plicitly considering the user comfort setting as a constraint.
The solution achieves an optimal balance between energy
cost and user comfort level.
• The two-step appliance commitment algorithm provides
adjustments to the optimal schedule to account for uncer-
tainties introduced by forecast errors in prices and con-
sumer consumptions.
• The algorithm is fast and robust. The central processing
Fig. 10. Comparison between the sorted forecast price and sorted actual price unit (CPU) time of the optimization results solved for a
(blue: forecast; red: actual). 24-hour EWH schedule on a PC (Intel Duo CPU@3.00
GHz) is 11 s on average. Because the linear-sequential-
optimization-enhanced, multiloop algorithm is fundamen-
tally an exhaustive search algorithm, the solution is optimal
and always solvable.
The algorithm can be used in home/building energy-manage-
ment systems to help household owners or building managers
to automatically create optimal load operation schedules based
on different cost and comfort settings and compare cost/benefits
among schedules.

Fig. 11. Hot water temperatures (blue: real-time adjustment; red: day-ahead REFERENCES
schedule) (the upper and lower dot lines represent the temperature limits).
[1] A. J. Wood and B. F. Wollenberg, Power Generation Operation and
TABLE III Control. New York: Wiley, 1996.
ELECTRICITY COST AS A FUNCTION OF DIFFERENT TEMPERATURE CONSTRAINT [2] S. Katipamula, D. P. Chassin, D. D. Hatley, R. G. Pratt, and D. J. Ham-
CORRESPONDING TO DIFFERENT COMFORT SETTING merstrom, Transactive controls: A market-based gridwise controls for
building systems PNNL-15921, Pacific Northwest Natl. Lab.. Rich-
land, WA, 2006.
[3] F. C. Schweppe, B. Daryanian, and R. D. Tabors, “Algorithms for a
spot price responding residential load controller,” IEEE Trans. Power
Syst., vol. 4, pp. 507–516, May 1989.
[4] B. Daryanian, R. E. Bohn, and R. D. Tabors, “Optimal demand-side
response to electricity spot prices for storage-type customers,” IEEE
Trans. Power Syst., vol. 4, pp. 897–903, 1989.
[5] S. H. Lee and C. L. Wilkins, “A practical approach to appliance load
control analysis: A water heater case study,” IEEE Trans. Power App.
Syst., vol. PAS-102, pp. 1007–1012, Apr. 1983.
[6] T. Ericson, “Direct load control of residential water heaters,” Energy
Policy, vol. 37, no. 9, pp. 3502–3512, Sep. 2009.
[7] C. H. K. Goh and J. Apt, Consumer strategies for controlling electric
water heaters under dynamic pricing, Carnegie Mellon Electricity In-
dustry Center , Pitsburgh, PA, Working Paper CEIC-04-02.
[8] A.-H. Mohsenian-Rad and A. Leon-Garcia, “Optimal residential load
control with price prediction in real-time electricity pricing environ-
ments,” IEEE Trans. Smart Grid, vol. 1, no. 2, pp. 120–133, Sep. 2010.
[9] M. A. A. Pedrasa, T. D. Spooner, and I. F. MacGill, “Coordinated
E. The Selection of Comfort Settings scheduling of residential distributed energy resources to optimize smart
home energy services,” IEEE Trans. Smart Grid, vol. 1, no. 2, pp.
A major benefit of load scheduling is that users can compare 134–143, Sep. 2010.
[10] N. Lu, J. H. Chow, and A. A. Desrochers, “Pumped-storage hydro-
the costs/benefits among different comfort settings. Table III de- turbine bidding strategies in a competitive electricity market,” IEEE
picts electricity payments associated with different upper and Trans. Power Syst., vol. 19, pp. 834–842, May 2004.
lower temperature settings of the EWH. As the temperature [11] J. C. van Tonder and I. E. Lane, “A load model to support demand man-
agement decisions on domestic storage water heater control strategy,”
band becomes wider or the temperature band is moved down- IEEE Trans. Power Syst., vol. 11, no. 4, pp. 1844–1849, 1996.
wards (e.g., from 156–165 F to 141–150 F), the cost of elec- [12] M. H. Nehrir, P. S. Dolan, V. Gerez, and W. J. Jameson, “Development
tricity is decreased. and validation of a physically-based computer model for predicting
winter electric heating loads,” IEEE Trans. Power Syst., vol. 10, no.
1, pp. 266–272, 1995.
V. CONCLUSION [13] N. Lu and D. P. Chassin, “A state queueing model of thermostatically
controlled appliances,” IEEE Trans. Power Syst., vol. 19, no. 3, pp.
This paper illustrates the formulation of an appliance com- 1666–1673, Aug. 2004.
[14] N. Lu, D. P. Chassin, and S. E. Widergren, “Modeling uncertainties
mitment problem and presents a two-step linear-sequential-op- in aggregated thermostatically controlled loads,” IEEE Trans. Power
timization-enhanced, multiloop algorithm to solve the problem. Syst., vol. 20, no. 2, pp. 725–733, May 2005.
The contributions of the paper are summarized as follows. [15] S. Katipamula and N. Lu, “Evaluation of residential HVAC control
strategies for demand response programs,” ASHRAE Trans., vol. 112,
• The appliance commitment algorithm provides a system- no. 1, pp. 1–12, 2006, PNNL-SA-45954, Pacific Northwest Natl. Lab.,
atic way to formulate an optimal load scheduling process Richland, WA.
DU AND LU: APPLIANCE COMMITMENT FOR HOUSEHOLD LOAD SCHEDULING 419

[16] R. G. Pratt and Z. T. Taylor, Development and testing of an equiva- [25] A. Tuohy, P. Meibom, E. Denny, and M. O’Malley, “Unit commitment
lent thermal parameter model of commercial buildings from time-se- for systems with significant wind penetration,” IEEE Trans. Power
ries end-use data, Pacific Northwest Lab.. Richland, WA, Apr. 1994. Syst., vol. 24, no. 2, pp. 592–601, May 2009.
[17] Z. T. Taylor and R. G. Pratt, “The effects of model simplifications on [26] R. E. Bohn, Spot pricing of public utility services, MIT-EL 82-031,
equivalent thermal parameters calculated from hourly building perfor- 1982, also available as MIT Energy Lab. Tech. Rep..
mance data,” in Proc. 1988 ACEEE Summer Study Energy Efficiency
Buildings, 1988, pp. 10.268–10.285.
[18] A. Pahwa and C. W. Brice, “Modeling and system identification of
residential air conditioning load,” IEEE Trans. Power App. Syst., vol.
PAS-104, no. 6, pp. 1418–1425, Jun. 1985.
[19] “2008 ASHRAE Handbook—HVAC Systems and Equipment” Amer. Pengwei Du (M’06) received the B.Sc. and M.Sc. degrees in electrical engi-
Soc. Heating, Refrigerating and Air-Conditioning Eng., Inc., 2008 neering from Southeast University, Nanjing, China, in 1997 and 2000, respec-
[Online]. Available: http://www.knovel.com/web/portal/browse/dis- tively, and the Ph.D. degree in electric power engineering from Rensselaer Poly-
play?_EXT_KNOVEL_DISPLAY_bookid=2396 technic Institute, Troy, NY, in 2006.
[20] D. J. Hammerstrom, J. Brous, T. A. Carlon, D. P. Chassin, C. Eu- He has worked at Pacific Northwest National Laboratory, Richland, WA,
stis, G. R. Horst, O. M. Järvegren, R. Kajfasz, W. Marek, P. Michie, since 2008. His research interests include distributed generation, power system
R. L. Munson, T. Oliver, and R. G. Pratt, Pacific Northwest Grid- modeling and analysis, and energy storage applications.
Wise Testbed Projects: Part 2. Grid Friendly Appliance Project, PNNL-
17079, Pacific Northwest Natl. Lab.. Richland, WA, 2007.
[21] J. Contreras, R. Espínola, F. J. Nogales, and A. J. Conejo, “ARIMA
models to predict next-day electricity prices,” IEEE Trans. Power Syst., Ning Lu (M’98–SM’05) received the B.S.E.E. degree from Harbin Institute of
vol. 18, no. 3, pp. 1014–1020, 2003. Technology, Harbin, China, in 1993, and the M.S. and Ph.D. degrees in electric
[22] Y.-Y. Hong and C.-Y. Hsiao, “Locational marginal price forecasting power engineering from Rensselaer Polytechnic Institute, Troy, NY, in 1999 and
in deregulated electricity markets using artificial intelligence,” IEEE 2002, respectively.
Proc.—Gener., Transm., Distrib., vol. 149, no. 5, pp. 621–626, 2002. She was with Shenyang Electric Power Survey and Design Institute from
[23] F. J. Nogales, J. Contreras, A. J. Conejo, and R. Espínola, “Forecasting 1993 to 1998. Currently, she is a senior research engineer with the Energy Sci-
next-day electricity prices by time series models,” IEEE Trans. Power ence & Technology Division, Pacific Northwest National Laboratory, Richland,
Syst., vol. 17, no. 2, pp. 342–348, 2002. WA. Her research interests include modeling and analysis of power system load
[24] R. E. Bohn, “Spot pricing of public utility services,” Ph.D. dissertation, behaviors, energy storage evaluation, renewable integration, climate impact on
Massachusetts Inst. Technol., Cambridge, 1982. power grids, and smart grid modeling and diagnosis.

You might also like