Copyright (C) IEEE - Paper accepted for publication at the

IEEE Transactions on Smart Grid (Power and Energy Society)

Author pre-print
1

Intelligent Energy Optimization for User Intelligible

Goals in Smart Home Environments
Fulvio Corno, Member, IEEE, and Faisal Razzak, Student Member, IEEE,

Abstract—Intelligent management of energy consumption is To cope with increasing energy needs the smart grid is a
one of the key issues for future energy distribution systems, promising infrastructure [6] which focuses on demand side
smart buildings, and consumer appliances. The problem can be management. It provides customers an ability to make in-
tackled both from the point of view of the utility provider, with
the intelligence embedded in the smart grid, or from the point of formed decisions about their energy consumption by adjusting
view of the consumer, thanks to suitable local energy management timing and quantity of their electrical usage [7], [8]. This
systems (EMS). Conserving energy, however, should respect the flexibility is enabled by pricing policies for electrical usage
user requirements regarding the desired state of the environment, over time [9], [10] and/or by dynamic demand scheduling
therefore an EMS should constantly and intelligently find the algorithms to optimize energy services in buildings [11], [12].
balance between user requirements and energy saving. The paper
proposes a solution to this problem, based on explicit high-level The smart grid infrastructure requires a two-way communica-
modeling of user intentions and automatic control of device tion through which appliances can be monitored and controlled
states through the solution and optimization of a constrained by a control center installed on the premises of the energy
Boolean satisfiability problem. The proposed approach has been provider which may lead to privacy and security issues [13].
integrated into a smart environment framework, and promising A complementary approach to energy management is the
preliminary results are reported.
local optimization of energy consumption using a locally
Index Terms—Building automation, Home automation, Do- installed Energy Management System (EMS) on the building
motic Effects, Electrical power optimization, Energy Manage- premises. Most EMS focus on making the consumer more
ment Systems, Energy optimization heuristic
aware of their electrical power usage and/or providing methods
to share this information with energy providers or third party
I. I NTRODUCTION application developers [14]–[17]. The research focuses on
different graphical illustrations of data related to consumed

I N the last decade, intelligence emerged as the basic com-

ponent to design modern home and building automation
systems. The term “intelligence” implies a provision of au-
energy to ease consumer comprehension [18], [19] and on
different tools and methodologies to share this data over the
web [14]. All these approaches need active user participation
tomated control over the buildings to solve interoperability in order to implement energy management strategies.
issues among devices from different vendors, to sense the This paper proposes a more automated approach, where the
environment, to provide context-aware services to the residents EMS may automatically act on appliances and control their
and to manage safety and security issues. Regardless of how power consumption, while satisfying user requirements about
ambitious and diverse the notions might seem, the research the current environment state. The presented approach is based
community has demonstrated the ability to achieve such goals on two pillars: the availability of an explicit model for the
using pilot projects [1]–[4]. In the past few years, energy smart home (such as provided by intelligent home gateways),
efficiency has become a key requirement for designing modern and the expression of user needs in a more abstract way. The
buildings and industries. The approaches in this regard not environment should be controlled by “user intelligible goals”
only rely on improving building structures and adopting more that represent the state of the environment perceived by the
efficient appliances but also aim at increasing user awareness user, on an abstract level. For example, the user may wish to
towards their energy usage. illuminate a room and this may be done by acting on lamps,
Energy efficiency has become one of the major concerns curtains and shutters in different ways. Therefore, the user
in today’s life, impacting almost all human activities, from achieves the effect of illuminating the room on an abstract
industrial and commercial, to leisure and vacation. According level.
to the statistics from the US Department of Energy and the The paper describes a novel approach to optimize the energy
European Union Energy Commission, global energy consump- usage in a building while achieving user intelligible goals. The
tion is likely to increase in the next decade, with residential main contributions of the paper are: adopting an explicit formal
and commercial buildings raising their aggregate figure to 20- modeling for user goals (based on the Domotic Effects mod-
40% of the total yearly consumption. If only electricity is eling framework); proposing an architecture that is compatible
considered, the consumption share allocated to buildings is with existing ambient intelligence solutions; describing an
suppose to increase up to 73%, evenly distributed between algorithm based on Boolean satisfiability (SAT) for computing
residential and commercial buildings [5]. the optimal solution and integrating the SAT algorithm with a
suitable heuristic in order to tackle combinatorial explosion.
The authors are with Politecnico di Torino, Dipartimento di Automatica e
Informatica, Torino, Italy The paper is divided into seven sections. A literature
e-mails: fulvio.corno@polito.it, faisal.razzak@polito.it overview is first provided in Section II. Section III describes
 2

the theoretical framework of the paper, i.e., the explicit high- III. U SER INTENTION MODELING
level modeling of user intention. The problem tackled in the To control the environment, an EMS needs to be aware
paper is then formalized in Section IV, while the approach of the structure of the environment, including a list of de-
adopted for optimization is described in Section V. Section VI vices, their locations, their available commands, and most
shows detailed results of a preliminary experiment. Finally, importantly their power consumption in their various operating
Section VII concludes the paper and highlights future work. states. For environment and device modeling the DogOnt [26]
ontology was adopted which is supported by the Dog [3]
II. R ELATED W ORKS
automation gateway. Device models include device features,
Hubert et al. [20] outlined that in order to realize the functionalities, commands and attainable states. In particular,
potential of the smart grid, optimization of energy usage is for each device the set of possible internal working states is
required at different consumer levels, i.e., residential, com- represented, and each state may influence power consumption.
mercial, industrial. In the domain of EMS, the literature on Information about power consumption of devices or device
optimizing the electrical power usage while achieving user classes is encoded in a simple light-weight DogPower ontol-
intelligible goals (in real time) is scarce but several researchers ogy, derived from the Energy Profile ontology proposed in
have addressed the energy optimization issue at different [14]. The combination of DogOnt and DogPower gives all
consumer levels. needed information about the environment.
Reference [21] advocates the need to build an intelligent User intentions, on the other hand, are modeled thanks to
decision support system which takes into account user prefer- the “Domotic Effects” framework, presented in this section. It
ences and behavior, and then tries to assist the user in reducing is an abstract high level modeling approach, which addresses
the energy consumption according to a dynamic notion of the concerns from both AmI designer and resident points of
price. A model is proposed that learns user preferences and view. Domotic Effects allow them to concentrate on high-
characteristics over time, and provides different alternatives for level goals that the environment should accomplish, without
efficient energy usage. However, the practical implementation worrying about the underlying device(s) fulfilling the goal.
of such model, i.e., how to integrate it with a home automa- Domotic Effects provide AmI designers with an abstraction
tion system and its feasibility was not discussed. Moreover, layer that enables defining generic goals inside the environ-
the model focuses on user’s preferences over devices rather ment, in a declarative way, and designing and developing
than on higher level intelligible goal. A similar approach intelligent applications. The high-level nature of Domotic
is proposed in [22]. Dynamic pricing and incentive pricing Effects also allows the residents to program their personal,
policies are adopted and advocated by many in the smart grid office or work spaces as they see fit: they can define different
community to optimize the energy usage [9], [23] but the user achievement criteria for a particular generic goal, by using
perspective is often missing. domain-specific operators.
Amir-Hamed et al. [24] propose optimization of residential
load control with price prediction in a Real-Time electricity
pricing environment. It minimizes the householder’s electricity A. Conceptual Modeling
costs by scheduling the operations of each appliance, subject The Domotic Effects modeling framework is organized in
to special needs of the user. The user perspective is modeled three tiers (Fig. 1), supporting the needs of designers and
as a waiting parameter in the scheduling problem, whose cost users. The core layer contains the basic class definitions for
increases with time. Therefore, each appliance operation is expressing Domotic Effects. Each Domotic Effect (DE) is
scheduled based on price of electricity and the value of the expressed as a function of existing device states or sensor
waiting parameter. values. Such function is expressed using a set of operators
Reference [25] proposed a system model that uses game that can be extended or modified by the AmI designer. The
theory to design a energy consumption scheduling game AmI layer encodes the set of operators defined or customized
among consumers to address demand side management. It by the AmI designer, depending on the application domain.
considers a single energy source and multiple consumers. Finally, the Instance layer represents the specific Domotic
The consumers automatically coordinate among each other Effects being defined in a specific environment.
to find optimal energy consumption on an hourly basis. The The choice of operators, and their associated numerical
scheduling problem is modeled over a set of consumers and domains and ranges, are customized depending on the domain
could face scalability issues when the number of consumers of application requirements. For example, in monitoring and
increases. This technique also lacks the description of model- control applications, Domotic Effects express logical values (in
ing consumer requirements. Boolean algebra), and the AmI designer would define Boolean
One potential weakness of all above proposals is that they operators to construct effects. On the other hand, in energy
focus entirely on minimizing energy consumption and ignore management applications, Domotic Effects would represent
other environment aspects, especially the user’s perspective. power and energy values, and AmI designers would define
The Ambient Intelligence (AmI) community has addressed real-valued aggregation operators.
such aspects in the domain of smart environments. Often A DE can be a Simple Effect (SE) if it depends upon a single
missed is the point that the EMS inside a building will be device being in a particular state or a sensor having some
part of a larger smart environment system, providing sensing, value range; otherwise it is a Complex Effect (CE) and depends
actuation and user interaction. upon combinations of other DEs (both simple and complex). A
 3

Instance layer
House 1 House 2 C. Representing Power with Domotic Effects
SecureHome Each device, in each operating state, consumes some amount

User
Ventilation of electrical power1 , thas is represented as a real-valued Simple

AmI Designer
TVScenario
Effect P (s), P : S(d) → ℜ+ .
The instantaneous power consumed by the whole environ-
AmI layer
ment is therefore represented as a Complex Effect P : G →
Boolean Real ... ℜ+ aggregating all individual power measurements:
X
P(g) = P (s(d)) (1)

Framework
HVAC Lighting
d∈D

Core layer
D. Domotic Effect Enforcement
Core Domotic Effect Classes For Boolean valued domotic effects, the user can request the
system to enforce particular domotic effects. “Effect Enforce-
ment” addresses the problem of finding at least one configura-
Figure 1. Domotic Effect Modeling Framework
tion g that satisfies the user request and using the automation
system to bring the home devices into that satisfying state.
The user request R is defined as a subset of the declared
modular “DogEffects” ontology provides a formal knowledge Boolean domotic effects: R ⊆ I. In simple terms, the user
base for describing DEs, and is organized in a structure that request R is the subset of DE i that the user wants to be
corresponds to the three tiers modeling framework. active (true) at a given instant.
Users can define several domotic effects, based on the Satisfying user requests amounts to solving the Boolean
operators defined for the environment. At any instant, each DE function FR (g) defined as:
has a (Boolean or real) value associated with it. For example, Y
a user may describe a Boolean Domotic Effect corresponding FR (g) = DE (2)
to the generic goal of lighting up a room, and this goal may DE ∈R
be reached by acting, in different ways, on lamps, curtains, To address the problem, the user request is transformed into
and window shutters in that room, possibly by taking in to a Boolean satisfiability problem (SAT). SAT is the problem of
account external conditions. determining if the variables of a given Boolean expression
can be assigned in such a way as to make the expression
evaluate to true. Each domotic effect defined in the DogEffects
B. Domotic Effect Formalization ontology is mapped to a Boolean variable. The functionality
of each effect operator defined in the AmI layer is mapped in
Given an intelligent ambient managing a set of devices D, terms of a Boolean sub-expression in the SAT problem. The
each device d is associated with a set of allowed states S(d); Boolean expressions for all complex domotic effects present in
depending on the nature of the device, states may be discrete the user request R are recursively constructed and conjuncted.
(e.g., {On, Off} for a lamp) or continuous (e.g., [0, 100] for The Boolean expressions are solved using a SAT solver, which
a volume knob). During system evolution, the actual state of if satisfiable gives the values (true or false) of all the variables.
each device is a time-dependent function s(d, t) ∈ S(d). As simple domotic effects represent the terminal nodes of the
The whole environment possesses a global state space G, expressions, the values corresponding to their variables give
represented us the device configurations that will satisfy the user request.
Q by the Cartesian product of all device state spaces:
G = d∈D S(d), thus defining a global environment state
g ∈ G. IV. P ROBLEM S TATEMENT
Formally, a Domotic Effect DE is defined as a function of Given the definitions in the previous sections, the goal of
the global state space: DE : G → V, where V is an application- the paper is to compute the minimum value of P(g), while
dependent value space. For Boolean application domains, satisfying the user request R. This correspond to a constrained
V = {0, 1}. A Simple Effect SE is a function that considers optimization of P(g) subject to the Boolean constraint FR (g).
the state of only one device, SE : S(d) → V; such function In this paper, the basic SAT-based approach for effect en-
is time-dependent since it depends on s(d, t). An operator forcement has been extended to find a solution with minimum
op is a function op : V N → V, where N represents the power consumption. Since the set of possible solutions may
number of operands of the specific op. A Complex Effect CE be extremely large, a suitable heuristic is proposed to get a
is represented by a couple (op, (DE 1 . . . DE N )) composed satisfactory low-power solution in acceptable CPU times.
of an operator name op and a list of Domotic Effects DE i Energy management techniques should in fact respond in
whose values are used as operands. Such function is also time- near real-time (NRT), by acting on a time scale compara-
dependent. ble with user requests and device state change frequencies.
A set I contains all domotic effects defined for an environ- 1 in this paper active instantaneous power is considered, although the
ment, i.e., I = {DE 1 , DE 2 . . . DE M }. modeling approach can be trivially extended to other electrical properties
 4

A. Heuristic
A novel power minimizing heuristic is proposed to deter-
mine in near real-time a configuration that consumes minimal
electrical power and satisfies the user’s request. Since the
heuristic is called only when the solution space is large (> Tc ),
this degree of freedom is exploited by trying to switch off
appliances that have the highest electrical power consumption.
Forcing a device to be switched off reduces the size of the
solution space, but it might render the SAT problem infeasible.
Therefore a greedy approach was adopted which tries to force
all the involved devices off, one by one, starting from the
highest-consuming SE. Those SE that render the problem
infeasible are kept free in the SAT problem. The others
Figure 2. Architecture of proposed approach are forced off. There is no guarantee that the configuration
received after applying the heuristic has the minimum power
consumption. There might be cases in which the combination
Normally, the computational delay should be less than a few of small power consuming devices in total consumes more
seconds. than the device with high power consumption, but such
conditions are rare and the experiments (Section VI) prove the
configuration with minimum power value is ususally achieved.
V. P ROPOSED A PPROACH Algorithm 1 shows the overall steps taken to find the optimized
configuration, and Algorithm 2 details the greedy procedure
To minimize power P(g) subject to user-requested domotic used to simplify the SAT problem.
effects FR (g), a Domotic Effect Optimizer module is devel-
oped (Fig. 2). The Domotic Effect Optimizer receives a user Algorithm 1 Overall approach
request and transforms it into a SAT problem, that is solved to SAT = SAT problem derived from FR (g)
find valid configuration(s). The number of configurations may if (solvable(SAT)) then
if (num solutions(SAT)> Tc ) then
be zero or more. If zero, the user request is not satisfiable.
SAT = Heuristic Algorithm (SAT)
Otherwise, a configuration with minimum power consumption end if
needs to be determined. device states = solve (SAT)
An exhaustive enumeration approach can be adopted, in end if
which each valid configuration is checked for its total power
consumption value P(g) and the one with minimum value
is enforced on the environment. However, the enumerated Algorithm 2 Heuristic Algorithm (SAT)
approach becomes computationally expensive and practically sorted SE = sort( all SE, decreasing power )
infeasible if the number of configurations is too large. for all ( SE in sorted SE ) do
To guarantee near real-time (NRT) execution, the number SAT’ = SAT ∩ ( SE=false )
if (solvable(SAT’)) then
of configurations returned by the SAT solver is compared
SAT = SAT’
with an experimentally-tuned configurable threshold Tc that end if
roughly corresponds to the number of configurations that may end for
be enumerated in one second. If the number of configurations return SAT
is lower than Tc , then exhaustive enumeration is fast enough to
achieve NRT responsiveness. Otherwise, a heuristic is applied
to guarantee results in NRT, even if the absolute optimum is
no longer guaranteed. VI. E XPERIMENTAL EVALUATION
The complete approach is highlighted in Fig. 2. At startup, To prove the validity of the proposed approach and measure
the Domotic Effect Optimizer queries the DogEffects and the performance of the proposed heuristic, a set of experiments
DogPower ontologies to get all the domotic effects and their were carried out. The “Domotic Effect” modeling framework
associated (device and power) information. The Domotic Effect was developed and integrated with Dog2.1 [3] as a new
Optimizer transforms the user request for particular values of Domotic Effect Optimizer bundle running in the Dog OSGi
domotic effects in to a correct set of Boolean equations and framework.
constraints, constructing a SAT problem. Then it feeds the A complete house environment was simulated, whose do-
SAT problem to a SAT solver. For our current implementation, motic structure was modeled as an instance of DogOnt ontol-
the Sat4j [27] solver is used. Based on the set of Boolean ogy. A new test bundle was developed to test the approach
equations, the Sat4j solver determines (if possible) the total and the proposed heuristic. The house environment contains
number of configurations that satisfy the set of Boolean 1500 user-defined Domotic Effects. These DE correspond to
equations. generic goals like securing or illuminating the house.
 5

The experiments have been run on a standard personal exceeds the configuration threshold Tc . For such cases,
computer with a quad-core Intel i5 processor and 4GB of the time to determine a configuration with minimum
RAM. power consumption exceeds the NRT requirements, or
is marked as unknown. Unknown refers to cases in
which the number of configurations exceed 100, 000. The
A. Use Cases
enumeration approach is practically infeasible in such
In the experiments, all possible combinations of six use cases and the proposed heuristic must be applied.
cases were enforced on the environment one after another.
These use cases I were Secure Home, BathRoom Illumination,
Home Illumination, Afternoon Lunch, Isolated Kitchen and
Morning Wakeup scenarios. The “Secure Home” use case B. Results
secures all the exit points of the house, i.e., all exit doors
and windows. This use case comprises many DEs providing To demonstrate the applicability and results of the proposed
the ability to secure different rooms of the house. This can heuristic theAbove Threshold cases were focused, only, since
be used in case of emergency, theft, robbery or fire etc. the exhaustive enumeration approach is sufficient for the Zero
The “BathRoom Illumination” combines small use cases that Configuration and Below Threshold cases, that have been
represent alternative ways to illuminate the bathroom. The dropped from the subsequent tables.
“Home Illumination” requires that all the rooms of the house Table II compares the time taken by the exhaustive enu-
are illuminated. Illumination can be either natural or artificial. meration approach against the time taken by the heuristic
The “Afternoon Lunch” deals with the daily routine of cooking described in Section V-A to determine a configuration with
lunch inside the kitchen. The “Isolated Kitchen” use case minimal power usage. The Enumeration Solution Time column
represents isolating the kitchen from the rest of the house represents the time (in milliseconds) taken by the enumeration
during cooking hours; this scenario does not consider the approach. The Heuristic Solution Time column represents the
energy spent for cooking, since that action is not automated. time (in milliseconds) taken by the heuristic. The Result col-
The “Morning WakeUp” use case maps a typical scenario umn reports the comparison, i.e., Solved, Good, or Responsive.
when a resident wants to perform a sequence of activities The case is “Solved” when the heuristic is able to find a
after waking up in morning, like illuminating the bedroom, configuration with minimal power consumption in NRT while
the kitchen and the bathroom, switching off the gas heater the enumeration approach is infeasible. The “Good” cases
inside the bedroom, switching on the kitchen television and mean that the heuristic solution is faster than enumeration,
the bathroom radio. while the “Responsive” label means that the heuristic solution
Since |I| = 6, there were 26 = 64 possible user requests, is slower but still well inside NRT.
or 63 if the trivial R = ∅ is omitted, where no domotic effect Table III shows the comparison of the computed power
is enforced. consumption values between the enumeration and the heuristic
Table I shows the total number of configurations and the approaches. The Enumeration Power Value column shows the
time taken by the exhaustive enumeration approach to find the minimum electrical power (Watt), when it can be exhaustively
total number of configurations, compute their power consump- computed. The Enumeration Est. Power Value column shows
tion and determine the configuration with minimum power the estimated minimum electrical power (Watt) found after
consumption. The first 6 columns report which use cases are 100, 000 iterations; this value is useful only as a comparison,
enforced (1) or not (0) by the user. The time is calculated in since the involved CPU time is unrealistic. The column Heuris-
milliseconds. When the number of configurations were very tic Power Value shows the power value (Watt) of the config-
large, the enumeration was stopped at 100, 000. uration found by the heuristic. The Result column shows our
For the application of the heuristic optimization, the prob- observations, i.e., Better, Poor, or Equal. In the “Better” cases
lems that require more that one second to be enumerated were the heuristic was able to find a configuration that consumes
selected. From the analysis of the computation times in Table I, less than the configuration found by the enumeration approach.
it is evident that these cases can be selected by choosing a The “Equal” label shows cases in which the heuristic was able
threshold value Tc equal to 150. to find the configuration that consumes minimum electrical
From Table I, three types of cases are observed. They are: power in a shorter time than the enumeration approach. Only
1) Zero Configurations: It refers to the case when the Sat4j two cases are marked with “Poor”, where the heuristic was
solver cannot find a configuration satisfying the user’s not able to find the minimum power, but this happened for
request, which means that the current combination of use infeasible cases, only, where no practical alternative approach
cases can not be enforced together. is available.
2) Below Threshold: It refers to the cases when the total The size of the search space seems also to influence the
number of configurations satisfying the user’s request are effectiveness of the heuristic procedure: for example, the first
less than Tc . In such cases, the enumeration approach row in Table III puts very few constraints over device states,
is sufficient to determine in NRT a configuration with and the heuristic is usable to find a good solution, while the
minimum power consumption and enforce it. second row adds some constraints (i.e., Isolated Kitchen), and
3) Above Threshold: It refers to the case when the total the narrower search space allows to find a better solution. The
number of configurations satisfying the user’s request same applies to rows 5 and 6.
 6

Table I
E NUMERATION APPROACH STATISTICS
Secure Home BathRoom Illumination Home Illumination Afternoon Lunch Isolated Kitchen Morning Wake Up No. Of Configurations Time (ms)
0 0 0 0 0 1 32 220
0 0 0 0 1 0 3 16
0 0 0 0 1 1 32 56
0 0 0 1 0 0 3 18
0 0 0 1 0 1 32 65
0 0 0 1 1 0 3 13
0 0 0 1 1 1 32 73
0 0 1 0 0 0 >100000 unknown
0 0 1 0 0 1 0 10
0 0 1 0 1 0 >100000 unknown
0 0 1 0 1 1 0 14
0 0 1 1 0 0 >100000 unknown
0 0 1 1 0 1 0 15
0 0 1 1 1 0 >100000 unknown
0 0 1 1 1 1 0 13
0 1 0 0 0 0 16 94
0 1 0 0 0 1 32 111
0 1 0 0 1 0 48 65
0 1 0 0 1 1 32 67
0 1 0 1 0 0 48 74
0 1 0 1 0 1 32 86
0 1 0 1 1 0 48 84
0 1 0 1 1 1 32 58
0 1 1 0 0 0 >100000 unknown
0 1 1 0 0 1 0 11
0 1 1 0 1 0 >100000 unknown
0 1 1 0 1 1 0 11
0 1 1 1 0 0 >100000 unknown
0 1 1 1 0 1 0 11
0 1 1 1 1 0 >100000 unknown
0 1 1 1 1 1 0 9
1 0 0 0 0 0 192 534
1 0 0 0 0 1 0 12
1 0 0 0 1 0 48 97
1 0 0 0 1 1 0 16
1 0 0 1 0 0 48 95
1 0 0 1 0 1 0 19
1 0 0 1 1 0 48 113
1 0 0 1 1 1 0 14
1 0 1 0 0 0 2304 5100
1 0 1 0 0 1 0 18
1 0 1 0 1 0 576 1285
1 0 1 0 1 1 0 16
1 0 1 1 0 0 576 1424
1 0 1 1 0 1 0 15
1 0 1 1 1 0 576 1392
1 0 1 1 1 1 0 12
1 1 0 0 0 0 768 1421
1 1 0 0 0 1 0 13
1 1 0 0 1 0 192 394
1 1 0 0 1 1 0 13
1 1 0 1 0 0 192 416
1 1 0 1 0 1 0 50
1 1 0 1 1 0 192 417
1 1 0 1 1 1 0 13
1 1 1 0 0 0 2304 4650
1 1 1 0 0 1 0 39
1 1 1 0 1 0 576 1215
1 1 1 0 1 1 0 30
1 1 1 1 0 0 576 1195
1 1 1 1 0 1 0 9
1 1 1 1 1 0 576 1387
1 1 1 1 1 1 0 13

C. Discussion From Table II, it can be seen that our proposed heuristic was
able to solve all cases in NRT, even where the total number
In our experiments, a total of 63 iterations were performed, of configurations made the enumeration approach infeasible.
corresponding to each possible R defined over an environment Most cases took around 1 second to be solved by the heuristic.
with over 1500 declared DEs. Two performance comparisons By observing the results, it can be stated that the proposed
were measured between the proposed heuristic and the enu- approach is feasible for integration with intelligent building
meration approach. systems.

• the comparison of time taken by the approaches to Table III compares the power values of the configuration
compute the best solution to the user’s request (Table II). obtained using the enumeration and the heuristic approach.
• the consumed electrical power by the enforced settings In cases where the total number of configurations were less
of domotic effects (Table III). than 100,000 it can be seen that the proposed heuristic always
 7

Table II
C OMPARISON OF SOLUTION TIME ( MILLISECONDS ) BETWEEN THE ENUMERATION APPROACH AND THE PROPOSED HEURISTIC APPROACH

BathRoom Home Enumeration

Secure Afternoon Isolated Morning No. Of Heuristic
Illumina- Illumina- Solution Time Result
Home Lunch Kitchen Wake Up Solution Solution Time
tion tion (ms)
0 0 1 0 0 0 >100000 unknown 556 Solved
0 0 1 0 1 0 >100000 unknown 743 Solved
0 0 1 1 0 0 >100000 unknown 689 Solved
0 0 1 1 1 0 >100000 unknown 1123 Solved
0 1 1 0 0 0 >100000 unknown 829 Solved
0 1 1 0 1 0 >100000 unknown 463 Solved
0 1 1 1 0 0 >100000 unknown 760 Solved
0 1 1 1 1 0 >100000 unknown 1172 Solved
1 0 0 0 0 0 192 534 506 Good
1 0 1 0 0 0 2304 5100 708 Good
1 0 1 0 1 0 576 1285 662 Good
1 0 1 1 0 0 576 1424 1299 Good
1 0 1 1 1 0 576 1392 907 Good
1 1 0 0 0 0 768 1421 443 Good
1 1 0 0 1 0 192 394 972 Responsive
1 1 0 1 0 0 192 416 578 Responsive
1 1 0 1 1 0 192 417 1252 Responsive
1 1 1 0 0 0 2304 4650 2358 Good
1 1 1 0 1 0 576 1215 1296 Responsive
1 1 1 1 0 0 576 1195 1018 Good
1 1 1 1 1 0 576 1387 1371 Good

Table III
C OMPARISON OF COMPUTED POWER VALUE BETWEEN THE ENUMERATION APPROACH AND THE PROPOSED HEURISTIC APPROACH
BathRoom Home Enumeration
Secure Afternoon Isolated Morning No. Of Enumeration Heuristic
Illumina- Illumina- Est. Power Result
Home Lunch Kitchen Wake Up Solution Power Value Power Value
tion tion Value
0 0 1 0 0 0 >100000 N/A 4047.02 5411.29 Poor
0 0 1 0 1 0 >100000 N/A 3355.93 2763.87 Better
0 0 1 1 0 0 >100000 N/A 4728.43 4136.37 Better
0 0 1 1 1 0 >100000 N/A 4728.43 4136.37 Better
0 1 1 0 0 0 >100000 N/A 3408.39 5411.29 Poor
0 1 1 0 1 0 >100000 N/A 2961.98 2763.87 Better
0 1 1 1 0 0 >100000 N/A 4334.48 4136.37 Better
0 1 1 1 1 0 >100000 N/A 4334.48 4136.37 Better
1 0 0 0 0 0 192 0 N/A 0 Equal
1 0 1 0 0 0 2304 2583.16 N/A 2583.16 Equal
1 0 1 0 1 0 576 2763.87 N/A 2763.87 Equal
1 0 1 1 0 0 576 4136.37 N/A 4136.37 Equal
1 0 1 1 1 0 576 4136.37 N/A 4136.37 Equal
1 1 0 0 0 0 768 175.88 N/A 175.88 Equal
1 1 0 0 1 0 192 1146.36 N/A 1146.36 Equal
1 1 0 1 0 0 192 2518.86 N/A 2518.86 Equal
1 1 0 1 1 0 192 2518.86 N/A 2518.86 Equal
1 1 1 0 0 0 2304 2583.16 N/A 2583.16 Equal
1 1 1 0 1 0 576 2763.87 N/A 2763.87 Equal
1 1 1 1 0 0 576 4136.37 N/A 4136.37 Equal
1 1 1 1 1 0 576 4136.37 N/A 4136.37 Equal
 8

finds the configuration with the absolute minimum electrical [6] S. Lukovic, V. Congradac, and F. Kulic, “A system level model of
power value. On the other hand, the cases in which the possible integration of building management system in smartgrid,”
Engineering. Complexity in, pp. 58–60, 2010.
number of configurations exceeds 100,000, the heuristic was [7] B. Davito, H. Tai, , and R. Uhlaner, “The smart grid and the promise of
able to quickly solve all of them, and in most of the cases it demand-side-management,” McKinsey & Company, Tech. Rep., 2011.
was able to find a configuration that consumed less electrical [8] P. Palensky and D. Dietrich, “Demand side management: Demand
response, intelligent energy systems, and smart loads,” Industrial In-
power, compared to an (inapplicable) enumeration approach. formatics, IEEE Transactions on, vol. 7, no. 3, pp. 381–388, 2011.
Hence the experiments prove the feasibility of the complete [9] H. Sui, Y. Sun, and W. Lee, “A demand side management model based
approach as well as highlighting the robustness of the proposed on advanced metering infrastructure,” in Electric Utility Deregulation
and Restructuring and Power Technologies (DRPT), 2011 4th Interna-
optimization heuristic. tional Conference on. IEEE, 2011, pp. 1586–1589.
[10] A. Faruqui, R. Hledik, and J. Tsoukalis, “The power of dynamic pricing,”
The Electricity Journal, vol. 22, no. 3, pp. 42–56, 2009.
VII. C ONCLUSION [11] I. Koutsopoulos and L. Tassiulas, “Challenges in demand load control
for the smart grid,” Network, IEEE, vol. 25, no. 5, pp. 16–21, 2011.
This paper tackles the minimization of power consumption [12] M. Pedrasa, T. Spooner, and I. MacGill, “Coordinated scheduling of
from the point of view of individual buildings or homes. Smart residential distributed energy resources to optimize smart home energy
environments may be equipped with an energy management services,” Smart Grid, IEEE Transactions on, vol. 1, no. 2, pp. 134–143,
2010.
system that is able to intelligently control the activation [13] P. McDaniel and S. McLaughlin, “Security and privacy challenges in
of devices and to minimize power accordingly, taking into the smart grid,” IEEE Security & Privacy, pp. 75–77, 2009.
account the varying requirements of the users. The approach [14] D. Bonino, F. Corno, and F. Razzak, “Enabling machine understandable
exchange of energy consumption information in intelligent domotic
exploits the degrees of freedom that are available when the environments,” Energy and Buildings, vol. 43, no. 6, pp. 1392 – 1402,
users express their requirements at a higher level, in a user- 2011.
intelligible way, rather than directly controlling the state of [15] N. Fumo, P. Mago, and R. Luck, “Methodology to estimate building
energy consumption using energyplus benchmark models,” Energy and
each device. Buildings, vol. 42, no. 12, pp. 2331–2337, 2010.
The Domotic Effects modeling framework that has been [16] L. Pérez-Lombard, J. Ortiz, and C. Pout, “A review on buildings energy
presented effectively enables users to easily express their needs consumption information,” Energy and Buildings, vol. 40, no. 3, pp.
394–398, 2008.
at a higher level, by means of a Boolean formalization of [17] M. Weiss, F. Mattern, T. Graml, T. Staake, and E. Fleisch, “Handy feed-
the Domotic Effects enforcement and a SAT problem. The back: connecting smart meters with mobile phones,” in Proceedings of
Boolean problem has been extended to minimize power con- the 8th International Conference on Mobile and Ubiquitous Multimedia,
ser. MUM ’09. New York, NY, USA: ACM, 2009, pp. 15:1–15:4.
sumption, in near real-time, while satisfying user requirements, [18] Y. Riche, J. Dodge, and R. A. Metoyer, “Studying always-on electricity
and a heuristic algorithm has been proposed to find satisfactory feedback in the home,” in Proceedings of the 28th international con-
power results while respecting timing constraints. ference on Human factors in computing systems, ser. CHI ’10. New
York, NY, USA: ACM, 2010, pp. 1995–1998.
The extensive results reported on a case study show the [19] S. Roberts, H. Humphries, and V. Hyldon, “Consumer preferences
feasibility and the robustness of the approach, making it for improving energy consumption feedback,” Centre for Sustainable
suitable for adoption in smart environments. The proposed Energy, Tech. Rep., 2004.
[20] T. Hubert and S. Grijalva, “Realizing smart grid benefits requires energy
approach can be extended to include further constraints like optimization algorithms at residential level,” in Innovative Smart Grid
reducing the number of state changes to conserve the life-time Technologies (ISGT), 2011 IEEE PES, jan. 2011, pp. 1 –8.
of the appliances, or taking into account the energy needed to [21] O. Sianaki, O. Hussain, T. Dillon, and A. Tabesh, “Intelligent decision
support system for including consumers preferences in residential en-
switch between states (e.g., for mechanically actuated devices). ergy consumption in smart grid,” in Proceedings of the 2010 Second
Currently the work is being done towards a better integration International Conference on Computational Intelligence, Modeling and
of the approach in the Dog gateway open source distribution, Simulation, 2010, pp. 154–159.
[22] H. Zhang, X. Xia, and J. Zhang, “A residential energy and power
and on devising intuitive user interfaces to monitor and control conservation system utilizing an optimization model,” in AFRICON,
the environments through the Domotic Effects paradigm. 2009. AFRICON ’09., sept. 2009, pp. 1 –6.
[23] T. Kim and H. Poor, “Scheduling power consumption with price
uncertainty,” Smart Grid, IEEE Transactions on, vol. 2, no. 3, pp. 519
R EFERENCES –527, sept. 2011.
[24] A.-H. Mohsenian-Rad and A. Leon-Garcia, “Optimal residential load
[1] D. J. Cook, J. C. Augusto, and V. R. Jakkula, “Ambient intelligence: control with price prediction in real-time electricity pricing environ-
Technologies, applications, and opportunities,” Pervasive and Mobile ments,” Smart Grid, IEEE Transactions on, vol. 1, no. 2, pp. 120 –133,
Computing, vol. 5, no. 4, pp. 277–298, 2009. sept. 2010.
[2] D. Cook, M. Youngblood, E. Heierman III, K. Gopalratnam, S. Rao, [25] A. Mohsenian-Rad, V. Wong, J. Jatskevich, R. Schober, and A. Leon-
A. Litvin, and F. Khawaja, “MavHome: An agent-based smart home,” Garcia, “Autonomous demand-side management based on game-
in Pervasive Computing and Communications, 2003.(PerCom 2003). theoretic energy consumption scheduling for the future smart grid,”
Proceedings of the First IEEE International Conference on. IEEE, Smart Grid, IEEE Transactions on, vol. 1, no. 3, pp. 320 –331, dec.
2003, pp. 521–524. 2010.
[3] D. Bonino, E. Castellina, and F. Corno, “The dog gateway: enabling [26] D. Bonino and F. Corno, “Dogont - ontology modeling for intelligent
ontology-based intelligent domotic environments,” Consumer Electron- domotic environments,” in Proceedings of the 7th International Con-
ics, IEEE Transactions on, vol. 54, no. 4, pp. 1656 –1664, november ference on The Semantic Web, ser. ISWC ’08. Berlin, Heidelberg:
2008. Springer-Verlag, 2008, pp. 790–803.
[4] C. Ge, Y. Li, X. Zhi, and W. Tong, “The intelligent stb-implementation [27] D. Le Berre and A. Parrain, “The Sat4j library, release 2.2 system de-
of next generation of residential gateway in digital home,” in Pervasive scription,” Journal on Satisfiability, Boolean Modeling and Computation,
Computing and Applications, 2007. ICPCA 2007. 2nd International vol. 7, pp. 59–64, 2010.
Conference on. IEEE, 2007, pp. 256–261.
[5] Department of Energy, USA, “2008 buildings energy data book,” Build-
ings Technologies Program Energy Efficiency and Renewable Energy,
Tech. Rep., 2009.