Electric Power Systems Research 81 (2011) 21552163

Contents lists available at ScienceDirect

Electric Power Systems Research

journal homepage: www.elsevier.com/locate/epsr

A novel energy service model and optimal scheduling algorithm for residential
distributed energy resources
Michael Angelo A. Pedrasa a, , Ted D. Spooner b , Iain F. MacGill b
a
b

Electrical and Electronics Engineering Institute, University of the Philippines, EEEI Building, Velasquez St., Diliman, Quezon City 1101, Philippines
University of New South Wales, Sydney, Australia

a r t i c l e

i n f o

Article history:
Received 10 July 2010
Received in revised form 21 June 2011
Accepted 24 June 2011
Available online 23 July 2011
Keywords:
Energy services
Distributed energy resources
Home automation
Electricity tariff
Particle swarm optimization
Co-evolutionary optimization

a b s t r a c t
We propose a novel decision-support tool that aims to optimize the provision of residential energy
services from the perspective of the end-user. The tool is composed of a novel energy service model and
a novel distributed energy resources scheduling algorithm. The proposed model takes into account the
time-varying demand and benet that end-users derive from different services, and assigns the benet
to the energy that realizes the service. The scheduling algorithm determines how distributed energy
resources available to the end-users and under their control should be operated so that the net benet of
energy services is maximized based on the energy service models, and their technical characteristics and
capabilities. The scheduling is a challenging optimization problem; hence, a heuristic simulation-based
approach based around cooperative particle swarm optimization is used. The paper presents a case study
where this decision-support tool is used to optimize the provision of desired energy services in a smart
home that includes a number of controllable loads, energy storage and photovoltaic generation.
2011 Elsevier B.V. All rights reserved.

1. Introduction
A prominent manifestation of economic progress in our modern
society is the consumption of energy services. Therefore, the welfare of citizens relies heavily on the reliability of commercial energy
supply most especially that of electricity. The current electricity
industry is designed and has evolved with the primary purpose of
generating and making available the required electricity demand
regardless of magnitude and location within the transmission and
distribution networks. There is also a huge imbalance in investment decisions: on a per kWh basis, the demand-side makes a
larger investment with respect to the consumption of energy services than that of the supply side with respect to generating and
making electricity available. However, decision-making is centralized and focused on the optimal operation of the generation and
transmission sectors, while the consumers are treated as passive
participants and price-takers in the market.
The projected increase in demand for energy services and the
concern on depleting fossil fuels and climate change introduce
modern challenges to the aging electricity infrastructure. It has
been widely perceived that there is a need to transform the cur-

Corresponding author. Tel.: +63 2 9252958; fax: +63 2 9252957.

E-mail addresses: mapedrasa@up.edu.ph, mpedrasa@gmail.com
Pedrasa).
0378-7796/$ see front matter 2011 Elsevier B.V. All rights reserved.
doi:10.1016/j.epsr.2011.06.013

(M.A.A.

rent electricity industry into a form that would support sustainable

energy service consumption. This may be achieved by formally
incorporating the provision of energy services to the operation of
the industry; that is, the industry should focus on the provision of
energy services instead of just making electricity available at all
power receptacles.
A path toward this transformation is the utilization of technical instruments like energy efcient practices and equipment,
distributed energy resources (DER) such as embedded generators, storage, and controllable load, and by employing distributed
decision-making to manage these resources. The transformation
may also be achieved by using economic instruments like dynamic
electricity tariffs in the form of real-time pricing, peak demand
charges, and feed-in tariff, among others [1]. Several electric utilities have already adopted some of these instruments and have
demonstrated promising potential.
Numerous nancial and technical benets can be derived from
energy efciency and DER [24]. However, knowing the nature of
these benets may not be enough to persuade consumers, retailers
and distribution utilities to actively invest in these practices and
technologies. The uptake may be stimulated by providing the consumers with decision-making instruments that could enable them
to evaluate the potential impacts of installing these resources, and
to make operational decisions once these resources are acquired.
Several utilities have already demonstrated that demand response
through price-based signals can reduce peak electricity demand.

2156

M.A.A. Pedrasa et al. / Electric Power Systems Research 81 (2011) 21552163

However, potentially more peak and cumulative demand reduction can be achieved if the consumers are enabled by decentralized
and intelligent decision-making tools.
This paper aims to provide an avenue for consumers to participate in the decision-making process with respect to the optimal
operation of the electricity industry. In particular, this paper
presents a decentralized energy service decision-support tool that
a. consumers can use to simulate and to assess the potential
impacts of adopting energy efcient measures and installing
DER,
b. consumers can use to make intelligent decisions with respect to
the operation of their controllable DER, and
c. incorporates user preferences in terms of perceived benets and
exibility with respect to their required services.
We have described the decision-support tool and demonstrated
its capabilities using a simple example in an earlier paper [5]. In this
paper, we investigate its application to a more complex scheduling
problem involving a smart home with more DER and being subject
to a broader range of electricity tariff structures.
2. The energy services decision-support tool
This paper presents a decision-support tool that consumers can
use to optimize the acquisition of their energy services. The tool
consists of an approach for modeling energy services and a DER
scheduling algorithm. The energy service models are based upon
end-users putting value to the benet derived from their various
energy services requirements. Using these models, the scheduling
algorithm (or scheduler) maximizes the net benet derived from
these energy services by proposing a strategy for how to operate
the available DER.
2.1. Modeling energy services
Energy services are energy forms, commodities and processes
from where end-users ultimately appreciate and derive the value
of commercial energy carriers like electricity and gas [6]. Some
of the common forms of energy services are space conditioning,
water heating, illumination, information processing and communications. The demand for an energy service depends on several
factors, most notable are occupancy patterns, end-user preferences
and habits, time of day and day of week, and season of the year.
End-users put value to the energy services they use, not to the
amount of electricity the equipment that deliver these services
consume [7]. The value of an energy service originates from the
comfort, convenience, products and prots it brings to the enduser. The end-user either perceives or may directly quantify the
benet of an energy service, or a combination of both. As example, a residential end-user may be willing to pay several dollars a
day for his house to be heated on cold nights (perceived value).
On the other hand, a semiconductor processing plant owner could
compute the thousands of dollars he would lose for each hour that
his plant is not running (directly quantied). The value of an energy
service to an end-user may depend on the time of the day, weather,
and social externalities, among others. To illustrate, the benet of
having a bright work area in an ofce building is certainly higher
during the day than during the night.
Energy services may be modeled by specifying the temporal
variation of their demand and perceived benets. The demand
may be described by specifying the required temporal changes to
a physical variable directly related to the service, like the hourly
temperature in a room or the volume of hourly consumption of hot

water. It may also be described by the actual energy consumption

of the end-use equipment used to deliver the service.
Benet may be assigned to an energy service in several different
ways. The simplest approach is to assign a xed value to a service
regardless of the amount of consumed electric energy or duration
of delivery. As an example, an end-user may perceive that the benet of having a warm shower during the winter season is $10/day.
To incorporate the total amount of energy that realizes a service
and its duration in the benet model, we can assign a monetary
value to each unit of energy that realizes the service for each hour
that it is utilized. As an example, the benet of hot water service
may be quantied by putting a dollar value to each unit of thermal energy contained by the hot water consumed every hour. The
thermal energy content of water may be thought of as the energy
equivalent of the hot water service.
This approach allows a distinction to be made between the
energy that realizes the energy service and the actual electricity
consumption of the end-use equipment. We are putting value on
the energy service itself and differentiating that value from the cost
of electricity consumption. The temporal variation of benet of a
service can therefore be represented by the hourly variation of the
value of each unit of energy equivalent.
It is generally difcult to determine the monetary equivalent
of an energy service. Some users can readily identify the benet
of a particular service while others cannot. Nevertheless, the proposed model enables the users to adjust the benets until they are
satised with the efciency of service provision.
Fig. 1 shows how the modeling approach can be used to model
the demand for and benet of hot water service in a house. The
demand may be represented by the hourly volume of hot water
consumption or by the required hourly amount of energy equivalent. The benet is represented by the hourly variation of the
monetary value assigned to each unit of energy equivalent. The
gure depicts that the residents put different levels of benet to the
hot water service and the value is not necessarily correlated with
the amount of hot water consumption.
The demand for shiftable and interruptible services like washing and pool pumping may not be represented as a change of some
variable or energy equivalent. In place of an hourly variation of
demand, a narrative description may be given. For example, the
demand for the dish washing service may be described as the dish
washing service requires 1 kW of electricity over a period of 1 h, and
the washer may run anytime between 8 PM and 11 PM. Benet is
modeled by assigning a positive benet value to the energy equivalent on those hours when the service should be delivered, and a
negative value (or cost) on those hours when the service should not
provided.
The relationship of the energy equivalent of a service, UES (t),
to the actual electric energy consumption, Pe (t), is determined for
each service to be modeled. The relationship heavily depends on
the physical processes occurring within the end-use equipment
and the service itself. The determination is straightforward if the
equipment instantly converts electricity to the end-use energy or
Demand for hot water service (liters or thermal kWh per hour)

time
Value of hot water service ($ / kWh)

12 MN

6 AM

12 NN

6 PM

time
12 MN

Fig. 1. An example of how to use the proposed energy service model to describe the
demand for and benet of hot water service in a house.

M.A.A. Pedrasa et al. / Electric Power Systems Research 81 (2011) 21552163

process. Examples of such equipment are light bulbs and some

appliances. If the efciency of the equipment in converting electricity to an energy service is , then,

2157

In some services, the conversion from electricity to service is not

instantaneous, that is, there is some form of storage involved. In
space heating service, for example, the room is not immediately
heated at the ick of the heaters switch. It would take several minutes or hours before the temperature reaches the desired level. In
the case study, we derive the relationship of the energy equivalents of space and water heating services to the electric energy
consumption of the heaters.

involved might require a powerful computer and long computation

times. On the other hand, simulation-based heuristic techniques
offer great potential. Although they do not guarantee the optimal
solution, they can produce feasible solutions within manageable
time frames. In the case study, we chose a heuristic optimization
method based on particle swarm optimization due to its effectiveness and ease of implementation.
The application of the decision-support tool to optimize energy
service provisions in larger systems (like a microgrid or a VPP)
requires a more powerful optimization algorithm. The schedules
may be derived using commercially available software with builtin mathematical programming packages like CPLEX [8] and GAMS
[9].

2.2. Mathematical formulation of scheduling DER

3. Particle swarm optimization

End-users can maximize the net benet from their energy services by planning (or scheduling) the operation of controllable DER.
The net benet is equal to the total benets derived from the services less the cost of electricity consumption. The benets derived
from the services are determined from the temporal variation of
service demand and value of the energy equivalents. The cost of
consumption is determined using the prevailing electricity tariff.
The DER scheduler then creates a strategy for how controllable DER should be operated. The strategy will be in the form of a
schedule, or a set of recommended actions at each interval of the
simulation horizon. The scheduler will also quantify the savings
incurred by operating the DER using the strategy, and this result
may be used for making investment decisions.
The scheduling is essentially an optimization problem that aims
to nd the DER operation schedule x that maximizes the benet of the energy services, less the cost of electricity consumption.
The mathematical formulation of the optimization problem is to
maximize

Particle swarm optimization (PSO) is a population-based search

technique that mimics how a group of simple particles could
achieve complex collective behaviors [10]. Each particle in the
swarm is a candidate solution, and their movement around the
solution space is affected by the best performing particle (Gbest)
and the best position (Pbest) they have ever achieved. The velocity
V (v1 , . . . , vi , . . . , vn ) and position P(p1 , . . ., pi , . . ., pn ) of a particle
in the (t + 1) th iteration is computed by

UES (t) = Pe (t).

S
T 


[ES,i (t) UES,i (t, x)] Cost

(1)

(2)

t=1 i=1

subject to the operational constraints of the DER. The summation

term is the benet derived from the energy services and Cost is
the cost of electric energy consumption. x is the DER operation
schedule, ES,i and UES,i are the benet and demand for the energy
equivalent of the i th service, T is the number of hours in the simulation period, and S is number of energy services. Cost includes
the energy and capacity charges, and payment for energy export if
feed-in compensation is available.
By enabling the end-user to assign different levels of benet
to different services, the decision-support tool can prioritize the
provision of high-valued services over lesser-valued services. Furthermore, the DER scheduler may suggest the curtailment of some
services if the cost of provision exceeds their benet. It is also possible to assign negative benets (or costs) therefore prohibiting the
acquisition of services during some periods of the day. This technique may be used to describe the exibility of shiftable loads. The
case study described in Section 4 demonstrates these features of
the decision-support tool.
If we desire that all services should be delivered, extremely high
benet values may be assigned to the energy equivalents in Eq.
(2). The net benet maximization problem is then transformed into
a cost minimization problem. The cost minimization approach usually applied to DER scheduling problems is therefore a subset of the
proposed net benet maximization approach described by Eq. (2).
Many techniques may be used to solve the complex mathematical optimization problem described by Eq. (2). Enumeration
methods like dynamic programming may be used if the optimal
schedule is required, however, the immense computational effort

vt+1
= vti + c1 rand() (ptGbest,i pti ) + c2 rand() (ptPbest,i pti ),
i
(3)

pt+1
= pti + vt+1
,
i
i

(4)

where is the contribution of the particles previous speed, c1 and

c2 are the pull of the global and personal bests, and rand () is a
uniform random number generator from 0 to 1.
The PSO procedure is executed by rst randomly initializing the
position and velocity of all particles, and choosing the initial global
and personal bests by evaluating the objective function. The initial
positions may be pre-processed to accelerate convergence or to
improve the probability of locating the solution. The velocity and
position of all particles are repeatedly updated using Eqs. (3) and
(4), and in each iteration, the tness of all particles are evaluated,
and the global and personal bests are updated if needed. The particles are made to search for a solution until the maximum number of
iterations is reached or a stopping criterion is satised. The global
best particle at the end of the simulation is taken as the solution
to the optimization problem. In some cases, the historical values of
the global best are recorded, and the best is chosen.
In [11], PSO was extended to solve binary-valued optimization
problems. In binary PSO, a position coordinate pi is either 0 or 1. The
are
velocity is also computed using Eq. (3), but the values of vt+1
i
constrained within a range [ Vmax , Vmax ]. To determine the value
of pt+1
, vt+1
is mapped to a probability using a sigmoid function,
i
i
Eq. (5), and the result is compared to a random number. That is, if
S(vt+1
) > rand(), then pt+1
= 1, otherwise, pt+1
= 0.
i
i
i
S(vt+1
)=
i

1
1 + exp(vt+1
)
i

(5)

The canonical version and binary versions of PSO has been shown
to be effective in generating near-optimal solutions to challenging
problems. Some of their applications to power system optimization are described in [1216]. It was demonstrated in [1719] that
PSO can generate near-optimal solutions within short computation
times when compared to other heuristic optimization methods.
Like any other optimization algorithm, PSO suffers from the
curse of dimensionality. This may be overcome by using cooperative co-evolution [20]. Cooperative PSO solves a high-dimensional

2158

M.A.A. Pedrasa et al. / Electric Power Systems Research 81 (2011) 21552163

optimization problem using a divide-and-conquer approach [21].

The vector to be optimized is divided into several component
vectors, and a swarm is created to optimize each component. To
determine the tness of a component vector, it is concatenated with
the global bests that the other swarms have generated so far. At the
end of the simulation, the solution is the concatenation of all global
bests.
PSO and its variations can offer signicant potential in locating near-optimal solutions in highly dimensional, non-convex and
non-continuous optimization problems. In PSO, all particles and
their histories contribute to the search, unlike in genetic algorithms
where poor particles are discarded. By remembering the personal
bests of all particles, all undiscovered good neighborhoods in the
vicinity of these personal bests may be explored. If the problem
is non-continuous, a good solution may be within close range of a
poor particle. That good solution may be found by that particle.
If the optimization problem is non-convex, several particles
could assume the global best distinction throughout the evolution.
Since the particles tend to follow the global best, the diversity of the
population may not degrade rapidly because the best particle they
will follow in the next iteration may be different from the current
best particle they are following. Toward the end of the evolution, as
determined by a certain convergence criteria like the incremental
change of the global best tness or when the maximum number
of iterations is reached, the particles also tend to cluster around
the best solution. Therefore, the neighborhoods of candidate solutions are also searched. The reported improvements by PSO in other
studies achieved through re-conguration and by forming hybrids
through combination with other optimization algorithms further
reinforce its status as a robust and effective algorithm.
The DER operation strategies in the case study are created using
cooperative PSO. The component vectors with real coordinates are
optimized using canonical PSO, while those with binary coordinates
are optimized using binary PSO.
4. A Smart home case study
4.1. Description of the case study
We explored the potential of the proposed decision-support tool
using a case study with a number of scenarios involving a smart
home with a range of controllable loads and distributed storage. The
DER are listed below and the operation of the rst four is scheduled.
1. Plug-in hybrid vehicle (PHEV). 5.9 kWh capacity, 3.0 kW maximum charging/discharging rate, 90% charging/discharging
efciency, may be discharged down to 30% of capacity, 0.1%
coulomb loss per hour.
2. Space heater. Provides the space heating service, maximum
heating power is 1.8 kW.
3. Storage water heater. Storage capacity is 80 L and the heating
element is rated 1.2 kW.
4. Pool pump. For pool maintenance. Rated 1.1 kW, may be run in
separate 2-h periods.
5. PV module. Peak output = 2.0 kW.
All energy services aside from PHEV battery charging, space
heating, hot water, and pool pumping are lumped together into an
aggregate must-run energy service. The must-run service includes
food storage, food preparation, illumination, and entertainment,
among others. The hourly electricity consumption of the must-run
service is shown in Fig. 2(a).
The desired hourly temperature is shown in Fig. 2(b). The residents prefer that the temperature is within 1 C from the desired
temperature. The residents leave at 8 AM and return at 5 PM so the

(a) 2.0
1.0
0
Desired

Outdoor

(b)
20
10

(c)

20
10
0
0

12
Time of day

18

24

Fig. 2. Hourly demand for the energy services: (a) Must-run service power consumption, in kWh; (b) Heating service: required hourly temperature, in C, the
outdoor temperature is also shown; (c) Hot water service: hot water consumption,
in liters.

desired temperature in that period is not indicated. The demand

for hot water is shown in Fig. 2(c).
The residents may or may not use the PHEV when they leave the
house. If they use the car, its batteries are fully discharged when
they return at 5 PM. If they leave the car at home, the scheduler
could use the car battery as a storage DER.
The residents have the following outlook on the services they
require: (1) the electric car should be fully charged by 8 AM; (2)
the space heating service is very important when they are at home
but they do not care about the temperature in the house while they
are away; (3) the hot water service is very important but has no
value during the times it is not consumed; (4) the must-run service
should run regardless of the prevailing price of electricity; and (5)
the pool pump should run at most 6 h anytime from 8 AM to 10 PM.
The simulation horizon is 1 day, divided into 24 1-h periods. The
simulation starts and ends at midnight. At each hour, the scheduler
determines (1) the charging or discharging rate of the PHEV battery,
(2) the heating power of the space heater, (3) whether the water
heater will be connected to the electricity supply or not, and (4)
whether to start a 2-h pumping period or not.
The DER scheduler formulates an operation strategy under different situations. Scenarios with different tariff arrangements are
explored, as well as the case when the car is, or is not, available. We
used combinations of the tariff structures shown in Table 1.
The value of scheduling is determined by simulating a baseline
case where the DER are manually controlled and comparing it to
cases where the DER are scheduled. In the baseline case, the residents use the car and charge the battery starting at 12 midnight.
The space heater is manually controlled to achieve the desired temperature. At 6 AM, the thermostat is adjusted to 21 C so that by 7
AM, the temperature is near the desired value. It is also lowered to
Table 1
Electricity tariff. These are the regulated retail rates in Sydney, Australia (ToU and
capacity) [22] and the state of Victoria (feed-in) [23] as of November 2009. The
Critical Peak Pricing (CPP) rates are those used by Energy Australia in their Strategic
Pricing Study [24].
Tariff (e (t), cap )

Rate

Time of Use (ToU), $/kWh

Peak (28 PM)
Shoulder (7 AM2 PM, 810 PM)
Off-peak (10 PM7 AM)

0.3564
0.1408
0.0814

Feed-in (net), $/kWh

0.60

Critical peak price (CPP), $/kWh

High alert (58 PM)

2.00

Capacity charge, $/kW (28 PM)

0.128186

M.A.A. Pedrasa et al. / Electric Power Systems Research 81 (2011) 21552163

Table 2
Mapping of perceived benet of services to monetary values.
Importance of energy service

Value ($/kWh)

Must-run
High
Medium
Dont Care
Expense

2.20
1.00
0.25
0.00
0.50

18 C at 10 PM, an hour before the desired reduction at 11 PM. It is

programmed to turn on at 3 PM with the thermostat set at 21 C so
the temperature is near the desired value at 5 PM. The pool pump
timer is set to run from 9 AM to 3 PM, while the PV output is highest and the residents are away. The water heater is permanently
connected to the ac mains.
4.2. Mathematical formulation
The mathematical formulation of the optimization problem,
based on Eq. (2), is to nd the DER schedule x = [xcar xheat xwater xpool ]
that maximizes
T


2159

to have; the negative value implies that the service does not benet the end-user, in fact, it will cost the end-user and its provision
reduces the net benet. The must-run services should be delivered
regardless of cost so the Must-run value (greater than the CPP rate)
is assigned to them.
The hourly monetary benet ES (t) assigned to the energy
equivalent of the services are shown in Fig. 3. These values are
based on the outlook of the residents to the services they require.
They prefer that the car is fully charged at 8 AM, so a High value
is assigned to the car charging service at 8 AM. They do not care
about the state of the batteries at other times so on those periods, a Dont Care value is assigned. A High value is assigned to the
space heating service when it is required. The residents do not care
about the indoor temperature from 8 AM to 5 PM so a Dont Care
value is assigned during that period. The value assigned to water
heating is High when the they are at home, and Dont Care when
they are away. Medium value is assigned to the pool pumping service from 8 AM to 9 PM. An Expense value is assigned from 10 PM
to 7 AM to prevent the pump from running. The must-run service should be provided so a Must-run value is assigned at all time
periods.

( ES,must-run (t)UES,must-run (t) + ES,car (t)UES,car (t, xcar ) + ES,heat (t)UES,heat (t, xheat ) + ES,water (t)UES,water (t, xwater ) + ES,pool (t)UES,pool (t, xpool ) e (t) Pe (t, x) )

t=1

(6)

cap max{Pe (14 : 20, x)}.

In Eq. (6), xcar and xheat are 24-element vectors which coordinates
are the hourly battery charging or discharging rates and the hourly
heating power of the space heater. xwater is also a 24-element vector
but with binary coordinates. The i th coordinate is equal to 1 if the
water heater is to be connected to the ac mains, otherwise it is 0.
The vector xpool contains the starting times and state of each 2-h
pumping period, formulated as:
xpool = [Start 1 Start 2 Start 3 State1 State2 State3 ].

(7)

The i th pumping period would run starting at Starti if Statei = 1.

If Statei = 0, the i th pumping period will not run. The last term in
the summation of Eq. (6) is the energy cost while the term outside the summation computes the capacity charge. Pe is the energy
imported from the grid.
The constraints of the optimization problem are as follows: (1)
the battery charging and discharging rates should not exceed the
maximum value; (2) the stored energy in the batteries should be
between 30% and 100% of capacity; (3) the space heating power is
non-negative and should be less than the maximum output; and
(4) the pool pumping periods should not overlap.
4.3. Energy service models
Benet is assigned to the services based on the residents perception of their importance. Table 2 maps the relative importance
of a service to a dollar value to be assigned to each kWh of energy
equivalent. The monetary equivalents were chosen arbitrarily
but are based on the time of use (ToU) and critical peak pricing
(CPP) rates shown in Table 1. The unit value of important services
(denoted as High) is higher than the peak electricity rate while
the unit value of medium-valued services (denoted as Medium) is
smaller than the peak but higher than the shoulder electricity rates.
These values imply that each kWh of energy equivalent of important services should be provided even during the peak period while
those of medium-valued services may be curtailed or postponed.
The Dont Care value is assigned to the energy equivalent of services that the residents do not care whether those are delivered or
not. The Expense value is assigned to services that they do not wish

We assumed that the space heating service is delivered if the

room temperature is within 1 C from the desired temperature.
Therefore, ES,heat (t) in Eq. (6) can only assume the High value if the
room temperature is within this range.
Physically based models are used to relate the energy equivalent (UES,i ) of the desired energy services to the hourly electricity
consumption (Pe,i ).
The energy equivalent of the car charging service, UES,car , is
the energy stored in the batteries. The stored energy at each hour
is derived from the charging schedule (xcar ), charging and discharging efciencies, and losses. The electricity consumption of the car
charging service, Pe,car , is equal to the charging rate of the batteries,
or
Pe,car = xcar .

(8)

The energy equivalent of the space heating service is the amount

of heat energy in the air so that its temperature is equal to the
desired value. This heat energy is equal to the heating load Q(t) of
the room which is also the heat owing out of the building shell.
According to [25],
Q (t) =

1
[ (t) out (t)] = UES,heat (t),
R des

(9)

where  des and  out are the desired and outdoor temperatures and
R is the thermal resistance of the building shell. The actual temperBattery charging:

High
Dont care
Space and water heating:

High
Dont care
Pool pumping:

Medium
Expense
Mustrun service:

Mustrun
Dont care

12

18

Time of day
Fig. 3. Benet assigned to each kWh of energy equivalent (ES (t)).

24

2160

M.A.A. Pedrasa et al. / Electric Power Systems Research 81 (2011) 21552163

Table 3
Description of case studies and simulation results.
Case number

Case description a , b

PHEV c

Export rate, $/kWh

Baseline: manual control of DER

ToU
B1
ToU
B2
$0.60
B3
B4
$0.00
B5
ToU
DER are scheduled:
S1
S1p
Yes
S2
S2p
Yes
S3 d
S3p d
Yes
S3h e
S3hp d
Yes
S4
S4p
Yes
S5
S5p
Yes
a
b
c
d
e

ToU
ToU
ToU
ToU
$0.60
$0.60
$0.60
$0.60
$0.00
$0.00
ToU
ToU

Energy cost, $

Capacity
charge

Energy export,
kWh

Energy
consumption,
kWh

CPP active

Yes

6.70
7.16
5.67
7.07
22.31

Yes
Yes

4.78
3.47
5.11
3.68
2.33
4.94
1.17
4.97
5.19
3.83
12.72
5.36

Yes

Yes
Yes

0.46

0.30
0.14

2.32
2.32
2.32
2.32
2.32

49.11
49.11
49.11
49.11
49.11

2.19
5.90
2.36
2.39
16.53
21.62
16.46
22.52
0.13
0.01
2.36
2.18

50.00
46.03
50.09
45.68
50.38
46.70
52.69
49.08
48.84
45.67
49.64
46.42

Blank entries mean No or Not Applicable.

See Section 4.1 for the discussion of the cases.
The PHEV is available as storage DER (indicated by the sufx p in the case number) sufx p in the case number.
The pool pump only runs for 4 h instead of 6.
Value of pool pumping service is increased from M edium to High (indicated by the sufx h in the case number.). The increase caused the pool pump to run for 6 h.

ature,  in , is determined by using it in Eq. (9) in place of  des , and the

resulting expression is combined with Eq. (10) to get Eq. (11). Eq.
(10) relates the change in temperature to the amount of thermal
energy introduced to the air.
d (t)
C in
= Pheat (t) Q (t),
dt

din (t)
1
= Pheat (t) [in (t) out (t)].
R
dt

(11)

The discrete-time equivalent of Eq. (11) using 1 h time steps is

in (t + 1) = in (t)e/ + RPheat (t)(1 e/ ) + out (t)(1 e/ ),
(12)
where  = 1 h and  = RC. The values used are R = 18 C/kW,
C = 0.525 kWh/ C, and the initial room temperature = 17 C. The
space heating thermal models described by Eqs. (11) and (12) were
adapted from [26]. We assumed that the electric energy consumption of the heater is entirely converted to heat so
Pe,heat = Pheat = xheat .

section then becomes part of the hot section and its volume is reset
to zero.
Tcold (t + 1) = Tinlet +

(10)

where C is the heat capacity of indoor air. Combining Eqs. (9) and
(10) yields
C

Capacity cost a ,$

(13)

For the storage water heater model, we assumed that the stored
water has two sections, hot and cold, having different temperatures. The hot section is already raised to the discharge temperature
and does not mix with the cold section. When hot water is discharged at the top, cold water comes in from the bottom and mixes
with the cold section.
The hourly decision made by the scheduler is whether to connect the heating element to the ac mains or not. If connected, the
cold section temperature at the end of the hour is estimated using
Eq. (14) if it takes more than 1 h to raise the temperature to the discharge value. However, if it takes less than 1 h, the heating element
is turned off, and the energy delivered by the heating element to
the cold section is computed using Eq. (15). The water in the cold

Pe,water (t) =

C0 [Tcold (t) Tinlet ]Vcold (t) + coil Pcoil Ploss

,
C0 [Vcold (t) + Vinlet (t)]
(14)

C0
{[Vcold (t) + Vinlet (t)]T Vcold (t)[Tcold (t) Tinlet ]} + Ploss .
coil

(15)

Eqs. (14) and (15) simply follow the law of energy conservation:
the increase in internal energy should be equal to the amount of
injected heat. Tcold and Vcold are the temperature and volume of
the cold section, and Tinlet and Vinlet are the inlet water temperature and volume. Pcoil = 1.2 kW and coil = 0.98 are the rated power
and efciency of the heating element. C0 = 1.167 103 kWh/L- C
is the specic heat of water. Ploss is the heat loss approximated using
the heater specications [27]. The inlet water should be raised by
T = 50 C.
The energy equivalent of the hot water service, UES,water , is the
thermal energy content of the discharged water:
UES,water (t) = C0 {VHW (t)(T ) + VCW [Tcold (t) Tinlet ]}.

(16)

VHW is the volume discharged from the hot section while VCW is the
volume discharged from the cold section.
Since the pool pump and the appliances providing the must-run
services are already acquired, their actual electric energy consumptions are assigned as the energy equivalents of the pool pumping
and must-run services. Therefore,
UES,pool = Pe,pool ,

(17)

UES,must-run = Pe,must-run .

(18)

The electricity consumption of the pool pump can be easily derived

from Eq. (7).
The hourly grid import Pe is computed by adding the electricity
consumption of all services less the PV output:
Pe = Pe,must-run + Pe,car + Pe,heat + Pe,water + Pe,pool PPV .

(19)

M.A.A. Pedrasa et al. / Electric Power Systems Research 81 (2011) 21552163

Heating service (desired and actual temperature, oC):

Heating service (desired and actual temperature, C):

25

25

20

20
15

15

DER operation strategies:

WH

SH

2.0

Grid import and PV generation, in kW:

3.0
0.0
6

12

2.0
0.0
1.0
0.0
1.0
0.0

6.0
3.0
0.0
3.0

6.0

100%
30%

PP

WH
PP

Batt

100%
30%
SH

Batt

DER operation strategies:

0.0
1.0
0.0
1.0
0.0

18

6.0

24

Grid import and PV generation, in kW:

12

Time of day

Heating service (desired and actual temperature, C):

25

20

20

15

15

DER operation strategies:

Batt

WH

SH

2.0

100%
30%

PP

SH

Batt

DER operation strategies:

100%
30%

WH

24

Heating service (desired and actual temperature, C):

PP

18

Time of day

25

0.0
1.0
0.0
1.0
0.0

2161

Grid import and PV generation, in kW:

6.0
3.0
0.0
3.0

2.0
0.0
1.0
0.0
1.0
0.0

Grid import and PV generation, in kW:

6.0
3.0
0.0
0

12

18

24

Time of day

12

18

24

Time of day

Fig. 4. Simulation results for cases S1, S1p, S3h, and S5p. Shown are the desired and actual indoor temperatures, DER schedules, and total grid import and PV output. The
DER operation schedules are described by the stored energy in the PHEV battery (Batt, % of full capacity), space heater power (SH, in kW), water heater power (WH, in kW),
and pool pump power (PP, in kW). The residents take the PHEV when they leave in (a) and (c), so the battery disappears at 8 AM, and returns discharged at 5 PM.

4.4. Scheduling of DER using PSO

We used cooperative PSO to solve the optimization problem
described by Eq. (6). Ten swarms are used to nd the strategy x:
three swarms optimize the charging of the battery (xcar ), three
swarms optimize the space heater power (xheat ), three swarms
optimize the water heater schedule (xwater ) and one swarm optimize the pool pumping schedule (xpool ). The 24 coordinates of xcar ,
xheat , and xwater are divided into three 8-coordinate vectors and a
swarm is used to optimize each 8-h vector. Canonical PSO is used
to determine the hourly charging and discharging rates of the battery, the heating power of the space heater, and the starting times
of the pool pumping periods (Starti in xpool ). The pool pump starting
times Starti are discretized by rounding them to the nearest hour.
Binary PSO is used to determine whether the water heater is to be
connected to the supply or not and if the i th pool pumping period
should run or not (Statei in xpool ).
The parameters used for canonical PSO are = 0.7298 and
c1 = c2 = 1.4962, and for binary PSO, = 1.0, c1 = c2 = 7.5, and
Vmax = 5.0. These sets of parameters were effective in generating
near-optimal solutions in [28] (for canonical PSO) and [15] (for
binary PSO). The swarms guided by canonical PSO have 50 particles while the swarms guided by binary PSO have 20 particles.

The simulation has 120 iterations. In each iteration, the order at

which the four DER schedules are evolved is chosen randomly but
the 3 swarms assigned to the heaters and batteries are evolved
in chronological order. The constraints are handled using a repair
algorithm [29], that is, the coordinates that violate the constraints
are corrected at each iteration.
4.5. Simulation results
The simulation results are summarized in Table 3. The case
description includes whether the PHEV is available from 8AM to
5PM as a storage DER, the energy export rates, whether capacity
charge is levied, and if CPP is active for the day. The table shows
the resulting energy and capacity costs, the amount of exported
energy, and the total energy consumption of the house. The baseline schedule is implemented under different tariff structures, and
the results are labeled B1 to B5. The cases where the DER are scheduled are labeled S1 to S5. Cases with the same number after S
and B have the same tariff structure. The sufx p indicates that
the PHEV is available as a storage DER from 8 AM to 5 PM, and
the sufx h indicates that the value of the pool pumping service
is increased from Medium to High from 8 AM to 9 PM. The DER
operation schedules of some cases are shown in Fig. 4. The g-

2162

M.A.A. Pedrasa et al. / Electric Power Systems Research 81 (2011) 21552163

ures show the actual indoor temperature, the state of charge of

the PHEV battery, the hourly energy consumption of the space
and water heaters and pool pump, and the hourly grid energy
import.
The results in Table 3 show that the cost of consumption is
reduced in all cases where the DER are scheduled when compared
to the baseline schedule with the same tariff structure. The peak
demand is also reduced when capacity charges are levied, and the
larger energy export is achieved when a premium energy export
rate is granted. The total energy consumption, however, is not necessarily reduced.
The DER schedules for case S1 (PHEV is not available, feed-in
rate is ToU, no capacity charges and no CPP) are shown Fig. 4(a). In
this case, all required services are delivered and the cost is reduced
by minimizing the operation of the DER during the peak period. The
cost of operating the space heater is minimized by pre-heating the
house during the shoulder period to keep it off during the rst 4 h
of the peak period and maintaining the temperature at the lower
level of the comfortable range. When the PHEV is available in (Case
S1p in Fig. 4(b)), the stored energy is released starting at 2 PM to
maximize the energy export during the peak period.
The DER operation in cases S2 and S2p are similar to that of
case S1. The capacity charge has no signicant effect since it is
applied during the peak period where DER operation is already
minimized.
In the case where premium feed-in tariff is available (cases S3,
S3p, S3h and S3hp), the strategy is to maximize the energy that
could be exported. This is achieved by minimizing DER operation
from 10 AM to 4 PM and exporting the excess PV output. The residents receive credit from the utility (negative costs) because of the
signicant grid export. The pump only runs for 4 h in S3 and S3p
because only Medium benet is assigned to the pool pumping service, whereas in S3h and S3hp, a High benet is assigned so it runs
for 6 h. In all four cases, a peculiar yet logical behavior is observed
from the PHEV batteries: it charged and discharged several times to
take advantage of the high net feed-in compensation. This battery
charging strategy is shown in Fig. 4(c).
Feed-in energy is not compensated in cases S4 and S4p so the
strategy is to minimize the energy export by scheduling the DER so
they would consume all PV energy output.
Critical peak pricing is active in cases S5 and S5p. The house is
pre-heated in both cases but the temperature drops below 20 C at
7 and 8 PM because the cost of providing the space heating service
exceeds the benet during these hours. In case S5p, the PHEV is
available so it is discharged during the CPP period. Fig. 4(d) depicts
these DER strategies.
The results of these simulations were able to provide the residents insights on how the DER may be operated in order to
maximize their net benets when subjected to different tariff
structures and perceived energy service values. The results also
demonstrate that cost reduction is a consequence of net benet
maximization, and in some cases, the residents earn some credit
in their electric bill by achieving signicant grid export. The DER
scheduler was able to suggest effective strategies by utilizing the
exibility of the pool pumping service, the heat storage properties
of air and hot water, and the energy storage capability of the PHEV
battery. Some of the formulated schedules, however, do not necessarily reduce the total energy consumption. On the cases when
the PHEV is not available from 8 AM to 5 PM, the increased energy
consumption is due to the heat leaking out of the house after it was
pre-heated during the shoulder rate period. On the cases where the
PHEV is available, the reduced consumption is due to the energy
discharged by the battery, and this energy is used to provide other
services. Finally, the scheduler can quantify the reduction in the
electric bill (or the increase in credits) if the suggested strategies
are followed.

The simulation for each case is run 10 times and the best schedule is chosen. The average simulation time for 10 runs is 320 s, using
Matlab 2008b on Windows XP running on a 2.0 GHz Pentium dual
CPU.
5. Conclusion
The conventional approach for acquiring energy services may
be optimized by recognizing that end-users put varying levels of
benet to different services at different times. Some services should
be provided almost regardless of cost, while some services may be
curtailed if the cost of provision exceeds their benet.
The energy service decision-support tool presented in this paper
was able to optimize the acquisition of energy services by enabling
the end-user to assign benet to energy services, and by scheduling
the operation of controllable DER. The energy service model may be
used to represent the temporal changes of benet and demand for
services. By assigning monetary value to each unit of energy that
realizes a service, or the energy equivalent, benet is assigned to
the service itself and actual service utilization is differentiated from
electric energy consumption. Furthermore, it enables us to incorporate the amount of energy needed to realize the service and the
duration of service provision to the assignment of benet. Using
a case study involving a smart home, we demonstrated how to
use the model to assign benet to services based on the perception of the end-user. We were able to show that some services may
be curtailed if the assigned benet is less than the cost of acquisition. In the case study, the decision-support tool recommends the
reduction of the pool pumping hours when premium feed-in tariff is
available, and curtails the space heating service when CPP is active
the indoor temperature dropped below the comfortable range
because the space heater is not operated during the CPP hours. We
were also able to show that undesired services were not provided
if a negative benet is assigned. This technique can hasten the creation of schedules because the handling of these type of constraints
is incorporated to the maximization of net benet. The energy service model, therefore, provides a unied approach for depicting
the temporal variation of perceived value of services, allowing the
possibility of service curtailment, and describing the exibility of
shiftable services.
We were also able to demonstrate through the case study that
the maximization of net benet could formulate DER schedules that
provide differently valued services at reduced costs. The DER scheduler was able to formulate effective operation schedules for four
DER under different tariff schemes and different values assigned
to services. The DER scheduling algorithm took advantage of the
exibility of some services, the availability of distributed storage
options, and the temporal variation of the cost of electricity to
create the DER operating schedules. The derived schedules were
also consistent with the objectives of demand response programs
implemented through effective tariff design. The results show that
electricity consumption is reduced during the ToU peak periods and
CPP hours, and the peak demand is reduced if peak demand charges
are levied. Grid export is also maximized if premium feed-in rates
are granted to the consumers.
The scheduling of DER is a complex mathematical optimization problem, and the chosen approach using cooperative particle
swarm optimization was able to produce efcient schedules within
manageable computation times.
Acknowledgments
Michael Angelo Pedrasa acknowledges the scholarship granted
by the University of the Philippines and the Philippines Department
of Science and Technology through the ERDT.

M.A.A. Pedrasa et al. / Electric Power Systems Research 81 (2011) 21552163

References
[1] J. Randolph, G.M. Masters, Energy for Sustainability: Technology, Planning, Policy, Island Press, Washington, 2008.
[2] A. Lovins, E. Datta, T. Feiler, K. Rabago, J. Swisher, A. Lehmann, K. Wicker, Small
is Protable: The Hidden Economic Benets of Making Electrical Resources the
Right Size, Rocky Mountain Institute, Snowmass, CO, 2002.
[3] J.A. Pecas Lopes, N. Hatziargyriou, J. Mutale, P. Djapic, N. Jenkins, Integrating distributed generation into electric power systems: a review of drivers, challenges
and opportunities, Electr. Power Syst. Res. 77 (2007) 11891203.
[4] M. Ilic, J.W. Black, M. Prica, Distributed electric power systems of the future:
institutional and technological drivers for near-optimal performance, Electr.
Power Syst. Res. 77 (2007) 11601177.
[5] M.A. Pedrasa, E.D. Spooner, I.F. MacGill, Improved energy services provision
through the intelligent control of distributed energy resources, in: 2009 IEEE
Bucharest Powertech Conference, Bucharest, Romania, July 2009.
[6] R. Haas, N. Nakicenovic, A. Ajanovic, T. Faber, L. Kranzi, A. Muller, G. Resch,
Towards sustainability of energy systems: a primer on how to apply the concept
of energy services to identify necessary trends and policies, Energy Policy 36
(2008) 40124021.
[7] G.M. Masters, Renewable and Efcient Electric Power Systems, John Wiley &
Sons, Inc., Hoboken, New Jersey, 2004.
[8] IBM ILOG CPLEX Optimizer: High-Performance Mathematical Programming
Engine. See also: http://www-01.ibm.com/software/integration/optimization/
cplex-optimizer/.
[9] General Algebraic Modeling System Home Page. See also: http://
www.gams.com/.
[10] J. Kennedy, R. Eberhart, Particle swarm optimization, in: Proceeding of the 1995
IEEE International Conference on Neural Networks, vol. 4, 1995, pp. 19421948.
[11] J. Kennedy, R. Eberhart, A discrete binary version of the particle swarm algorithm, in: Proceedings of the 1997 IEEE International Conference on Systems,
Man and Cybernetics, vol. 5, 1997, pp. 41044108.
[12] Y. Del Valle, G.K. Venayagamoorthy, S. Mohagheghi, J-C. Hernandez, R.G. Harley,
Particle swarm optimization: basic concepts, variants and applications in
power systems, IEEE Trans. Evol. Comput. 12 (2008) 171195.
[13] A.D. Yucekaya, J. Valenzuela, G. Dozier, Strategic bidding in electricity markets using particle swarm optimization, Electr. Power Syst. Res. 79 (2009)
335345.
[14] M. El-Telbany, F. El-Karmi, Short-term forecasting of Jordanian electricity
demand using particle swarm optimization, Electr. Power Syst. Res. 78 (2008)
425433.

2163

[15] M.A. Pedrasa, E.D. Spooner, I.F. MacGill, Scheduling of demand side resources
using binary particle swarm optimization, IEEE Trans. Power Syst. 24 (2009)
11731181.
[16] P. Reche Lopez, M. Gomez Gonzalez, N. Ruiz Reyes, F. Jurado, Optimization of
biomass fueled systems for distributed power generation using Particle Swarm
Optimization, Electr. Power Syst. Res. 78 (2008) 14481455.
[17] T. Niknam, A.M. Ranjbar, A.R. Shirani, B. Mozafari, A. Ostadi, Optimal operation
of distribution system with regard to distributed generation: a comparison
of evolutionary methods, in: Conference Record of the Fortieth IAS Annual
Meeting Industry Applications Conference, vol. 4, 2005, pp. 26902697.
[18] E. Elbeltagi, T. Hegazy, D. Grierson, Comparison among ve evolutionary-based
optimization algorithms, Adv. Eng. Inform. 19 (2005) 4353.
[19] S.P. Karthikeyan, K. Palanisamy, L.J. Varghese, I.J. Raglend, D.P. Kothari, Comparison of intelligent techniques to solve load economic dispatch problem with line
ow constraints, in: Proceedings of the IEEE International Advanced Computing
Conference, 2009, pp. 446452.
[20] Y. Liu, X. Yao, Q. Zhao, T. Higuchi, Scaling up fast evolutionary programming
with cooperative coevolution, in: Proceedings of the 2001 Congress on Evolutionary Computation, vol. 2, 2001, pp. 11011108.
[21] F. van den Bergh, A.P. Engelbrecht, A cooperative approach to particle swarm
optimization, IEEE Trans. Evol. Comput. 8 (2004) 225239.
[22] Energy Australia Regulated Residential Energy Rates. Sydney, Australia. See
also: http://www.energy.com.au/energy/ea.nsf/Content/NSW+Your+energy+
agreement+ResidentialYA.
[23] Department of Primary Industries: Victorian Feed-in Tariffs. Melbourne,
Australia. See also: http://www.dpi.vic.gov.au/dpi/dpinenergy.nsf/Home+
Page/EnergyHome+Page?open.
[24] C. Amos, Advanced tariff design. Workshop on Advance Metering Infrastructure, Sydney, Australia, 2006. See also: http://www.ceem.unsw.edu.au/
windworkshop/AMIworkshop31ChrisAmos.pdf.
[25] H. Sauer, R. Howell, W. Coad, Principles of Heating, Ventilating and Air Conditioning, ASHRAE, Atlanta, USA, 2001.
[26] L.D. Ha, S. Ploix, E. Zamai, M. Jacomino, Tabu search for the optimization of
household energy consumption, in: Proceedings of the 2006 IEEE International
Conference on Information Reuse and Integration, 2006, pp. 8692.
[27] Rheem Hot Water Manual. See also: http://www.rheem.com.au/manuals.asp?
view=commercial.
[28] F. van den Bergh, A.P. Engelbrecht, A study of particle swarm optimization
trajectories, Inf. Sci. 176 (2006) 937971.
[29] C.A. Coello, Theoretical and numerical constraint-handling techniques used
with evolutionary algorithms: a survey of the state of the art, Comput. Methods
Appl. Mech. Eng. 191 (1112) (2002) 12451287.