IEEE Communications Magazine - June 2017 - page 207

205
IEEE Communications Magazine • June 2017
0163-6804/17/$25.00 © 2017 IEEE
A
bstract
By recognizing patterns in occupants’ daily
activities, building systems are able to optimize
and personalize services. Established technolo-
gies are available for data collection and pattern
mining, but they all share the drawback that the
methodology used for data collection tends to be
ill suited for pattern recognition. For this research,
we developed a bespoke WSN and combined
it with a compact data format for frequent epi-
sode mining to overcome this obstacle. The pro-
posed framework has been evaluated with both
synthetic data from a smart home simulator and
with real data from a self-organizing WSN in a stu-
dent’s home. We are able to demonstrate that the
framework is capable of discovering sequential
patterns in heterogeneous sensor data. With cor-
responding scenarios, patterns in daily activities
can be deduced. The framework is self-contained,
scalable, and energy-efficient, and is thus applica-
ble in multiple building system settings.
I
ntroduction
From a technical point of view, a smart city tries
to improve the quality of life of its citizens in
terms of urban services by utilizing information
and communication technology (ICT), data min-
ing (DM), and other new technologies to improve
urban services. Patterns of daily behavior are a
decisive factor in many aspects of the urban envi-
ronment, including traffic, air quality, energy cost,
and so on. Considering that urban dwellers spend
approximately 90 percent of their lives indoors,
it is no surprise that buildings are responsible for
about two-thirds of all electrical energy consump-
tion. Making cities smarter begins indoors. Utiliz-
ing data collection to reveal occupants’ behavior
patterns enables data-driven decisions for smarter
buildings. Usage can be anticipated, thus reduc-
ing the consumption while improving the experi-
ence [1].
The data-driven decision for buildings is largely
based on data acquisition and data mining tech-
nologies. Wireless sensor networks (WSNs) are
widely used for data acquisition since wireless is
low cost and more flexible than wired solutions
[2]. Extensive research has been performed on
efficiency [3, 4] and mobility [5–7]. Since it is
impossible for the system designer to envision all
possible contexts beforehand, decision making
control systems largely rely on data mining and
machine learning techniques such as an artificial
neural network (ANN), a support vector machine
(SVM), a self-organizing map (SOM), a hidden
Markov model (HMM), and frequent pattern min-
ing (FPM)[8]. Artificial intelligence provides many
benefits for data gathering systems [9].
E
xisting
P
roblem
In spite of all of the studies conducted on data
acquisition and data mining for building systems,
no practical solution has been provided. There
are several reasons for this.
Too Complex:
Most of the research was con-
ducted in an experimental environment that
required expert installation, maintenance, and
upgrade of the system. Some mining algorithms
require extensive parameter settings that are not
intuitive and need professional and prior knowl-
edge of the environment. Supervisory algorithms
need additional training data that is hard to obtain
in a real world application.
Too Simple:
There are two main types of data
that can be recorded in a building system: numer-
ical (discrete and continuous sensor values) and
categorical (weather conditions: windy, snowy,
sunny, etc.). Unfortunately, most algorithms can
only address one type at a time. Although data-
types are interchangeable, additional parameters
and prior knowledge of the dataset are required.
Gap in Research Fields:
Established technol-
ogies are available for data collection and data
mining, but they all share the drawback that meth-
odology used for data collection tends to be ill
suited for purposes of data mining. WSN develop-
ers continue to improve efficiency, regardless of
the type of data and measuring frequency actually
needed. DM researchers focus on accuracy of the
algorithm, without regard for the data source or
collection efficiency.
S
olution
This research integrates data acquisition and data
mining techniques efficiently and practically to dis-
cover behavior patterns. We first propose a com-
pact data format that encompasses both sensor
data about spontaneous events and periodic envi-
Li Li, Xin Li, Zhihan Lu, Jaime Lloret, and Houbing Song
A
CCEPTED FROM
O
PEN
C
ALL
Sequential Behavior Pattern Discovery with
Frequent Episode Mining and
Wireless Sensor Network
By recognizing patterns in
occupants’ daily activities,
building systems are able
to optimize and person-
alize services. Established
technologies are available
for data collection and
pattern mining, but they
all share the drawback
that the methodology
used for data collection
tends to be ill suited for
pattern recognition. For
this research, the authors
developed a WSN and
combined it with a com-
pact data format for fre-
quent episode mining to
overcome this obstacle.
Li Li is with Southeast University; Xin Li is with Wuhan University; Zhihan Lu is with University College London;
Jaime Lloret is with University Politecnica de Valencia, Spain; Houbing Song is with West Virginia University.
Digital Object Identifier:
10.1109/MCOM.2017.1600276
1...,197,198,199,200,201,202,203,204,205,206 208,209,210,211,212,213,214,215,216,217,...228
Powered by FlippingBook