A Fuzzy Adaptive Comfort Temperature Model with

Grey Predictor for Multi-agent Control System of

Smart Building
Zhu Wang, Rui Yang, Lingfeng Wang, Robert C.
Green II

Anastasios I. Dounis
Department of Automation
Technological Educational Institute of Piraeus
Egaleo, Greece
aidounis@otenet.gr

Department of Electrical Engineering and Computer

Science
University of Toledo
Toledo, OH, USA
Lingfeng.Wang@utoledo.edu

preferences through a Graphical User Interface (GUI) [4]. For

indoor temperature control, users define their own comfort
zone by setting the parameters of minimum temperature and
maximum temperature [Tmin, Tmax]. Through the use of Particle
Swarm Optimization (PSO) the control system is able to
autonomously find the optimal set point that lies in the userdefined comfort zone. According to previous works regarding
the variation of indoor temperature [5]-[8], the indoor comfort
temperature is determined by a combination of the control
system and the variations in outdoor temperature. Other factors
including the business culture and metabolic rates of
individuals also have an impact when determining the most
comfortable indoor temperature.

Abstract In this paper a fuzzy adaptive comfort temperature

(FACT) model has been proposed for the intelligent control of
smart buildings. A multi-agent control system is applied for the
energy management and building operation. Particle Swarm
Optimization (PSO) is applied to optimize the set points based on
the comfort zone. Integrating a grey predictor to predict outdoor
temperature with the FACT model shows great promise in
systematically determining the customer temperature comfort
zone for smart buildings. With the application of the FACT
model and other intelligent technologies, the multi-agent control
system has successfully provided a high-level of temperature
comfort with low power consumption to customers in smart
building environments. Case studies and corresponding
simulation results are presented and discussed in this paper.

In this paper we introduce a fuzzy adaptive comfort

temperature (FACT) model integrated with a grey predictor
that enables the prediction of outdoor temperatures. When
applying the FACT model one important issue is the prediction
of meteorological parameters. This is an interesting and
challenging issue since the weather data is typically
complicated and disordered. Short-term prediction is used to
make use of the most recent data to forecast the future. This is
called local prediction. A Grey predictor is employed in this
model due to its good performance in local predictions despite
requiring a limited amount of past data to make these
predictions. As compared to methods that use only customerdefined comfort zones, the FACT model improves the
determination of the temperature limits systematically without
the intervention of users, achieves energy savings, and
enhances the intelligence of the building management system.
Managers are also provided an upper level of overreaching
controls in order to simplify the control process.

Keywords- Fuzzy logic; grey prediction; optimization; smart

building; energy efficiency

I.

INTRODUCTION

At the cutting-edge of modern building management

technologies, smart buildings are utilizing computers and
intelligent technologies to manage energy consumption and the
indoor environment autonomously. Smart buildings have
shown several advantages such as providing high-level comfort
to customers, minimizing power consumption, and reducing
impacts on the environment [1, 2]. In order to make the smart
building environmentally friendly, renewable energy sources
have been employed as the primary power supply. Through the
use of an intelligent control system a high-level of comfort and
power-efficiency should be easily achieved in a smart building
environment.
The indoor temperature, the indoor air quality, and the
indoor illumination are the most important comfort-impact
factors in smart buildings [3]. Our previous work [4] has
proposed a multi-agent control system with heuristic
optimization for the comfort and energy management of smart
buildings. Occupants are provided with a high level of
flexibility in order to participate in the control of system
behaviors. This means that they can set up their own

978-1-4244-7835-4/11/$26.00 2011 IEEE

The remainder of the paper is organized as follows. Section

II briefly reviews the multi-agent control system for smart
buildings. Section III represents the fuzzy adaptive comfort
temperature model including a detailed description of the grey
predictor. In section IV case studies and simulation results are
presented. The conclusion and future work are given in Section
VI.

728

II.

mathematical model of the comfort value that needs to be

maximized is defined as follows:

MULTI-AGENT CONTROL SYSTEM DESCRIPTION

A hierarchical, multi-agent technology is applied to the

control system in smart buildings. This design utilizes multiple
agents that are placed on different layers to achieve the control
goal of maximizing customer comfort while minimizing energy
consumption. Fig. 1 illustrates the structure of the multi-agent
control system. Based on the different functionalities all of the
agents are classified into four different layers. A switch agent
lies on the first layer; the second layer contains a central
coordinator-agent; the third layer is populated by three local
controller-agents that control three primary comfort impact
factors; and the fourth layer contains a load agent. The control
goal can be achieved through the coordination and cooperation
of all the agents.

Comfort = 1[1 (eT / Tset ) 2 ] + 2 [1 (eL / Lset ) 2 ]

+ 3 [1 (e A / Aset ) 2 ]

(1)

where Comfort is the overall customer comfort (in the range of

[0,1] ) and 1, 2, and 3 are the user-defined weighting factors
which indicate the importance of three comfort factors and
resolve the possible equipment conflicts. Customers can set
their own preferred values in different situations through GUIs.
All user-defined weighting factors fall into the range of [0,1]
and the sum of 1, 2, and 3 must be equal to one. Tset, Lset, and
Aset represent the set points of the temperature, the illumination,
and the air quality respectively. They can be obtained from the
particle swarm optimizer. The value for e is the error between
the measured value and set value. The measured value can be
obtained from the multiple, local controller-agents.
C. Particle Swarm Optimization (PSO)
Particle swarm optimization is first proposed as a selfadaptive and population based algorithm by Kennedy and
Eberhart in 1995 [9, 10]. It only has a few parameters needing
to be adjusted, and it is easier to escape from the local best
solutions and find the global best solution. PSO has turned out
to be a powerful tool to solve highly complex problems such as
large-scale non-linear optimization.

Figure 1. The structure of the multi-agent control system for smart building

Similar to other evolutionary technologies, in PSO a

population of particles is initialized first. Every potential
solution is seen as a particle in PSO, and each particle includes
both location vector l and the velocity vector v. Through
adjusting the velocity and memorizing the current location, the
particle moves towards its local best location pbest and finally
finds the global best location gbest in the search space.

A. Switch Agent
The switch agent is the bridge between the utility grid and
the smart building. It determines whether or not the building
should connect to or disconnect from the utility grid by
monitoring the current condition of the utility grid. As shown
in Fig. 1, the renewable energy source is the primary energy
supply and the utility grid is the secondary energy source.
When the smart building connects to the utility grid, the switch
agent determines and monitors the energy flow to ensure
maximum benefit for its inhabitants. The switch agent
purchases energy from the utility grid when the renewable
energy is insufficient and sells the redundant energy back to the
grid when the renewable energy production is larger than the
buildings energy consumption. The switch agent can
disconnect the smart building from the utility grid when
abnormalities occur. These include electrical disturbances
and unacceptably high power rates.

The update rules are presented as follows:

v ( k + 1) = v ( k ) + 1r1 ( pbest ( k ) l ( k ))
+ 2 r2 ( g best ( k ) l ( k ))
l (k + 1) = l (k ) + v(k + 1)

= max ( max min ) k n / k max

(2)
(3)
(4)

where is the inertia weight, max and min are the maximum
value and minimum value of the inertia weights; 1 and 2 are
two acceleration constants, r1 and r2 are two random numbers
from [0,1]; k is the iteration index, kn is the current iteration
number and kmax is the maximum number of iterations [11, 12].

B. Central Coordinator-agent
The primary task of the central coordinator-agent is to
coordinate the power dispatch and maximize customer comfort.
It determines the power dispatch to the load agent and multiple
local controller-agents in order to maximize the overall comfort
value according to the customers preference. This is done
through the consideration of the online energy production and
the outdoor environmental information. The proposed

The procedures of the PSO are listed as follows:

1) Randomly initialize a population of particles;
2) Evaluated the finess value to each particel according to
the objective function.
3) Calculate pbest and gbest for each particel;

729

Considering the different functionalities of buildings and the

variety of equipment they contain, customers should be given
the authority to manage the controllable loads based on their
own preferences. The load agent, when it is activated, is used to
decide the priority and the amount of load shed based on the
users preferences. More energy may be dispatched to the
comfort-related elements to maintain high-level comfort after
the load agent is working.

4) Adjust the velocity and location of each particle based

on the update rules (2)(3)(4);
5) Repeat 2) to 4) until the desired number of iteration is
achieved or the optimal solution is attained.
The Set-Point Particle Swarm Optimizer (SP-PSO) is
embedded in the central coordinator-agent to tune the set points
according to the outdoor information and the customer comfort
zone. The optimization goal is to maximize the objective
function defined in (1). In our previous work people are
provided with the flexibility to set their own comfort zones. In
this paper we provide a new comfort temperature model which
can automatically find the temperature comfort zone while
customers still have flexibility to fine tune the control system
by adjusting their degree of acceptability.

III.

FUZZY ADAPTIVE COMFORT TEMPERATURE (FACT)

MODEL

A. Adaptive Comfort Temperature Model

Beyond a simple logical algorithm, adaptive control is an
artificial intelligence technology which can be used for
empirical and judgmental information [6, 7]. The adaptive
models have recently been applied to define the indoor comfort
temperature as a linear regression which is related to outdoor
information. This can be expressed by an equation of the form:

D. Local controller-agents
Local controller-agents are implemented in three local
subsystems to control temperature comfort, visual comfort, and
air quality. Fig. 2 shows the structure of the local subsystems.

Tc = C + KTo

(5)

where Tc is the indoor comfort temperature (C), To is the

outdoor temperature (C) and C and K are constants. For
different buildings the constants C and K are found to be
distinct. For naturally ventilated buildings, which have heating
in the winter and are free-running (no cooling or mechanical
ventilation) in summer, the American Society of Heating,
Refrigerating, and Air-Conditioning (ASHRAE) standard (552004) adapts the model as follows:
Tc = 17.8 + 0.31To

(6)

Adaptive comfort temperature models are now included in

the Chartered Institute of Building Engineers (CIBSE) (2006)
guidelines and most recently the European Committee for
Standardization (CEN) standard EN 15251 has adopted the
following model:

Figure. 2 The structure of the local controller-agent

The local controller-agent employs a fuzzy controller to

determine the actual power to be used. It uses the dispatched
power from the central coordinator-agent, the error between
real environmental parameters, and the set points as inputs to
its fuzzy controller. Fuzzy rules are applied to calculate the
required power in uncertain circumstances. Then a comparison
is carried out between the required power calculated and the
adjusted power calculated by the central coordinator-agent in
order to determine the actual power. The actuators are driven
by the actual power used to control indoor environmental
parameters that influence users overall comfort level. The
actuators are auxiliary heating/cooling, electrical lighting, and
ventilation for controlling the thermal comfort, visual comfort,
and air quality. In this way the indoor environmental
parameters can be controlled by the corresponding actuators in
local subsystems.

Tc = 18.8 + 0.33To

(7)

Considering the effect of both heating and free-running

situations, the equations for indoor comfort temperature models
are different because the indoor temperature is decoupled from
the outdoor temperature by the heating control when the
heating is turned on. It has been shown that heating systems are
more likely to be on than off when running mean outdoor
temperature (Tmr) is less than 10C. In CIBSE 2007 and CEN
Standard EN 15252-2007 the equations linking comfort
temperature to outdoor temperature are:

E. Load Agent
The load agent controls all the interruptible loads. Any
equipment that has no direct connections to the three main
comfort factors is considered to be interruptible loads.

730

Tc = 18.8 + 0.33Tmr , Tmr 10 C (free-running)

(8)

Tc = 22.6 + 0.09Tmr , Tmr <10 C (heating on)

(9)

~
~
where ~ , and T are fuzzy numbers and

In [7], Peeters uses a new parameter named To,ref to define

the comfort temperature. The modified set of equations is
shown as:

~1 = (l 1 , m 1 , r 1 )
~

(18)
(19)

(10)

1 = (l 1 , m 1 , r 1 )

~2 = (l 2 , m 2 , r 2 )

(20)

Tc = 16.63 + 0.36To,ref , To, ref 12.5 C (free-running) (11)

2 = (l 2 , m 2 , r 2 )

(21)

~
To, ref = (lo , mo , ro )

(22)

~
Tc = (lc , mc , rc )

(23)

To,ref

T + 0.8Tk 1 + 0.4Tk 2 + 0.2Tk 3

= k
2 .4

Tc = 20.4 + 0.06To, ref , To, ref <12.5 C (heating on) (12)

where To,ref is the reference outdoor temperature (C), Tk is the
arithmetic average of todays maximum and minimum outdoor
temperature (C), Tk-1 is the arithmetic average of yesterdays
maximum and minimum outdoor temperature (C), Tk-2 is the
arithmetic average of the day before yesterdays maximum and
minimum outdoor temperature (C), and Tk-3 is the arithmetic
average of maximum and minimum outdoor temperature of
three days ago (C).

Equations (16) and (17) above are combined with Equation

(10) to form the FACT model to the multi-agent control system
of the smart building.
C. Fuzzy Arithmetic Operations

~
Assume two generalized triangular fuzzy numbers X1 and
~
~
~
X 2 where X 1 = (a1 , b1 , c1 ) and X 2 = (a2 , b2 , c2 ) . Using the
fuzzy arithmetic operators from [14] and [15] the following
equations can be given:

For air-conditioned buildings the correlation between

comfort temperature and outdoor temperature is given by [13]:
Tc = 18.6 + 0.16To

1) Fuzzy Numbers Addition

(13)

~
~
X1 X 2 = (a1 , b1 , c1 ) (a2 , b2 , c2 )
= (a1 + a2 , b1 + b2 , c1 + c2 )

As the primary goal of the smart building is to provide a

high comfort level to customers, a naturally ventilated building
cannot be the best choice for a smart building. We propose a
mixed building model which provides heat to the building in
the winter and cooling in the summer. The comfort temperature
correlated to the outdoor temperature for smart buildings has
been proposed using the following equations:

2) Fuzzy Numbers Multiplication

~
~
X 1 X 2 = (a1 , b1 , c1 ) (a2 , b2 , c2 )
= (a1 a2 , b1 b2 , c1 c2 )

Tc = 18.6 + 0.16To, ref , To, ref 18 C (air-conditioner on) (14)

Tc = 20.4 + 0.06To, ref , To, ref <18 C (heating on)

(24)

(25)

We
consider
three
fuzzy
~
~
= (l , m , r )
,
numbers
= (l , m , r )
~
and To, ref = (lo , mo , ro ) , where l, m, and r are real numbers.

(15)

B. Fuzzy Adaptive Comfort Temperature Model

Changing outdoor weather conditions, internal heat gains,
ventilation, and the preferences of users all influence the indoor
environment of a smart building leading to different comfort
profiles for different persons. The acceptable comfort
temperature regions are formulated by an upper and a lower
limit which defines the comfort temperature band. In this paper
a new method to determine the upper (Tmax) and the lower
(Tmin) bounds of the temperature band based on Equations (14)
and (15) is proposed. This method can be implemented using
the following fuzzy equations:

Utilizing the fuzzy arithmetic operators and , the

following fuzzy adaptive comfort temperature model is
developed:

~ ~
~
Tc = ~1 + 1 To,ref , To, ref 18 C (air-conditioner on) (16)

The user participates in the control process by defining the

that is the acceptability degree of comfort temperature. The
~
level determines the crisp set, that is, the lower Tc, min and the
~
upper Tc, max comfort temperatures in Equations (27) and (28).

~ ~
~
Tc = ~2 + 2 To ,ref , To, ref <18 C (heating on)

~
Tc = (lc , mc , rc )
~ ~
= ~ T

o, ref

(26)

= (l + l lo , m + m mo , r + r ro )
The relationship of these fuzzy numbers is lc=l+llo,
mc=m+mmo and rc=r+rro.

(17)

731

~
Tc, min = mc (mc lc )

(27)

~
Tc, max = mc + ( rc mc )

(28)

dx (1)
= x (1) (i ) + x (1) (i 1) = x ( 0) (i )
dt

And by approximating x(1) with [x(1)(i)+x(1)(i-1)], Equation

(30) can be rewritten as,

The acceptability degree, , will substantially change the

acceptance comfort zone. It is obvious that the bandwidth of
the comfort zone will extend when the acceptable degree, , is
changed.

x ( 0) (i ) + a y (1) (i ) = b

a
T
1 T
= ( B B) B Z n
b

y (1) (2)
(1)
y (3)
B=

#
(1)
y

( n)

(35)

x ( 0) (2)
( 0)
x (3)
Zn =
#
( 0)
x (n)

To weaken the randomness of the original raw data, the

original raw data x(0) is pre-processed and transformed into a
new sequence x(1) using the accumulated generating operations
(AGO).

1
#

and

Assume the original raw data series x(0) with n samples is

defined as x(0)=[ x(0)(1), x(0)(1), x(0)(2), .., x(0)(n)]. All values in
this data sequence are required to be positive. When negative
values appear in this sequence the absolute value of the
maximum negative data is added to all the data in the sequence
to make the data sequence positive.

x ( 0) (k ), i = 1,..., n

(34)

where

Local prediction forecasts of the future based on the most

recent data set are a type of time series. Suppose we have the
previous values of x from the time k-m to k-l, that is x(k-1), x(k2), , x(k-m). The next time interval values, x(k), can be
predicted by a grey model. The algorithm of this first order
grey model GM(1,1) is as follows [17].

(33)

The parameters a and b can be obtained by using the Least

Square Error Method in Equation (34)

D. Grey Predictor
Generally, a grey system means that the information
regarding the system is incomplete or uncertain. By using a
grey model that requires little previous data to perform a realtime forecast, the grey predictor has been successfully
employed in many areas. A first order linear dynamic grey
model, GM(1,1), is applied to make short-term predictions of
the future average outdoor temperature on a daily basis [16].

x (1) (i ) = AGO( x ( 0) (i )) =

(32)

(36)

The solution of equation (30) is an exponential function

with the initial condition x(1)(0)= x(0)(1). The predicted value
can be obtained as
b
b
x (1) (n + 1) = ( x ( 0) (1) ) e an +
a
a

(29)

(37)

k =1

Applying the inverse AGO to x (1) (n + 1) , we can have the

Therefore, the new raw data series is defined as x(1)=[

x(1)(1), x(1)(1), x(1)(2), .., x(1)(n)]..The x(1) sequence can be
modeled by a first-order differential equation that is defined as

dx (1)
+ ax (1) = b
dt

predicted
( 0)

(1)

for

x ( 0) (n + 1)

as

(1)

x (n + 1) = x (n + 1) x (n) . Usually the number of used

data in a grey model is rather small. Therefore, the grey model
is often used as a local prediction scheme. The grey model
prediction can be viewed as a curve fitting approach and is
employed in this study.

(30)

where a is the development coefficient and b is the grey control

variable. We now define y(1)(i) as the sequence obtained by
applying the MEAN operation to x(1) as seen in Equation (31).

y (1) (i ) = MEAN ( x (1) ) =

value

IV.

CASE STUDY AND SIMULATION RESULTS

Two sets of data of the daily average temperatures are

collected from [18] regarding the area around Toledo, OH in
the US. The first data set is a winter daily average temperatures
from Jan 1th to January 15th in 2011; the second data set is a
summer data set from August 1th to August 15th in 2010. By
applying a grey predictor we can forecast the average
temperature of the next day by analyzing the former three days
realistic data. The temperature in the first three days is used for

1 (1)
[ x (i ) + x (1) (i 1)], i = 1,..., n (31)
2

Suppose the sampling time is normalized as 1. From

Equation (29) we have

732

comfort temperature zones. The temperature is taken as the

most important comfort impact factor in the control of the
smart building. The user-defined weighting factors are set as
1=1 and 2=3=0. According to Equations (16) and (17) all of
the fuzzy members should be defined. Based on Equations (14)
and (15), we set the fuzzy member as:
~ = (l , m , r ) = (16.6,18.6,20.6)
(38)

prediction and the temperature for Day 4 through Day 15 is

forecasted. Table I and Table II show the comparison between
the predicted temperature and the realistic temperature.
TABLE I.

THE PREDICTED TEMPERATURE AND THE REALISTIC

TEMPERATURE IN WINTER
Winter data set
Date

Day 1

5.4

Day 2

-6.3

Day 3

-2.9

Day 4

-0.9

1.4

Day 5

-5.6

-8.1

Day 6

-3.4

-0.5

Day 7

-6.7

-8.9

Day 8

-8.5

-9.9

Day 9

-8.9

-9.2

Day 10

-4.4

1.8

Day 11

-4.6

-4.7

Day 12

-5.1

-5.5

Day 13

-7.1

-8.6

Day 14

-6.3

-5.4

Day 15

-2.8

1.9

TABLE II.

1 1

1 = (l 1 , m 1 , r 1 ) = (0.13,0.16,0.19)

(39)

~2 = (l 2 , m 2 , r 2 ) = (18.4,20.4,22.4)

(40)

2 = (l 2 , m 2 , r 2 ) = (0.03,0.06,0.09)

For T0,ref, we use Equation (10) to calculate the reference

temperature using the information from the grey predictor.
Considering that the reference temperature is related to the
temperature of the past three days, only 12 days (Day 4 Day
15) reference temperatures can be obtained out of the 15 days
of outdoor temperature information. The customer acceptability
degree is defined as =0.8. For comparison, the user defined
temperature comfort zones in both winter and summer periods
are set to [Tmin, Tmax] =[20,24 ](C) .
In the first case the predicted daily average temperature in
winter is used. The FACT model is applied to calculate the
comfort zone [Tmin, Tmax] and the set points are tuned by the PPSO algorithm. Fig. 3 and Fig. 4 illustrate the comfort zone and
the set point both with and without the FACT model. The
comfort zone after applying the FACT model is lower than
before. This demonstrates that the PSO has successfully found
the set points.

THE PREDICTED TEMPERATURE AND THE REALISTIC

TEMPERATURE IN SUMMER

(Aug.4-Aug.15) /C
Real
Predicted
temperature
temperature

Day 1

24.2

Day 2

23.5

Day 3

25.6

Day 4

25.7

25.8

Day 5

25.8

25.9

Day 6

23.4

21.2

Day 7

23.1

22.8

Day 8

24.5

25.9

Day 9

25.7

26.9

Day 10

26.8

27.9

Day 11

25.2

23.6

Day 12

24.6

24.0

Day 13

28.1

32.0

Day 14

26.3

24.6

Day 15

26.7

27.1

(41)

~
To, ref = (lo , mo , ro ) = (To, ref 2, To, ref , To, ref + 2)(C ) (42)

Summer data set

Date

(Jan.4-Jan.15) /C
Real
Predicted
temperature
temperature

Fig. 5 shows the energy saving advantage of the FACT

model. The power consumption for the local temperature
controller-agent is obviously reduced under the effect of FACT
model. The FACT model not only provides a more rational
comfort zone systematically to enhance customers comfort
level, but also reduces the power consumption of the building.
These advantages achieve the requirements of the control
system for smart building.
The second case uses the summer set data to control the
building and to observe the effect of the FACT model during
hotter weather. Fig. 6 and Fig. 7 are the comfort zones and set
points with and without FACT model in this case. Fig. 8 is the
comparison of the power consumption under these situations.

Based on the two data sets, two case studies are conducted
in this section by utilizing the FACT model to determine the

733

30

35

25
30

15

Upper Comfort Temperature

10

Low er Comfort Temperature

Set Points

Temperature(C)

Temperature(C)

20

Outdoor Temperature

25

20
Upper Comfort Temperature

Low er Comfort Temperature

15

Set Points

-5
-10

Outdoor Temperature
10

9 10 11
Time (Day)

12

13

14

15

9
10 11
Time (Day)

12

13

14

15

Figure 6. Comfort zone and set points with FACT model in summer

Figure 3. Comfort zone and set points with FACT model in winter

35

30
25

30

15

Upper Comfort Temperature

10

Low er Comfort Temperature

Set Points

Temperature(C)

Temperature(C)

20

Outdoor Temperature

25

20
Upper Comfort Temperature
Low er Comfort Temperature
Set Points

15

Outdoor Temperature

-5
10

-10

9 10 11 12 13 14 15
Time (Day)

9 10 11
Time (Day)

12

13

14

15

Figure 7. Comfort zone and set points without FACT model in summer

Figure 4. Comfort zone and set points without FACT model in winter
10

30
With FACT model

8
Power Consumption (KW)

Power Consumption (KW)

25
20
15
With FACT model
10

Without FACT model

Without FACT model

5
0

9 10 11 12 13 14 15
Time (Day)

9 10 11
Time (Day)

12

13

14

15

Figure 8. Power consumption for local temperature controller-agent with and

without FACT model in summer

Figure 5. Power consumption for local temperature controller-agent with and

without FACT model in winter

According to our current results, the FACT model can

provide high-level of comfort temperature in any outdoor
situation and significantly save energy in order to maximize the
customer benefit. It can be concluded that the FACT model is a
suitable method for smart building as it enhances the
intelligence and optimizes the entire building system.

734

V.

CONCLUSION

This study applied the FACT model with grey predictor to

the control system of a smart building. From the simulation
results, the appropriateness of this model is demonstrated and
discussed. The customers systematically attain a more
reasonable comfort temperature with less power consumption
when using the FACT model. In future studies similar adaptive
comfort models for visual comfort and air quality will be
developed in order to enhance the overall comfort level of the
smart building while consuming minimum energy.
REFERENCES
[1]
[2]

[3]

[4]

[5]

[6]
[7]

[8]
[9]
[10]

[11]

[12]

[13]

[14]
[15]

[16]

[17]
[18]

http://www.smart-buildings.com
J. Hang, J. Tian and H. Lin, Application of artificial neural network in
intelligent building, Proc. IEEE Int. Joint Conf. on Machine Learning
and Cybernetics, vol.7, pp. 4215 4220, Aug 2007
A.I. Dounis and C. Caraiscos, Advanced control systems engineering
for energy and comfort management in a building environmentA
review, Renewable and Sustainable Energy Reviews, vol.13, issue 6-7,
pp. 12461261, August September 2009.
Z. Wang, R. Yang, L. Wang and A.I. Dounis, Customer-centered
Control System for Intelligent and Green Building with Heuristic
Optimization, IEEE Power System Conference and Exposition,
Phoenix, AZ, March 2011
K. W. H. Mui and W. T. D. Chan, Adaptive comfort temperature model
of air-conditioned building in Hong Kong, Building and Environment,
vol.38, pp 837 852, June 2003
K. J. McCatney and J. F. Nicol, Developing an adaptive algorithm for
Europe, Energy and Buildings, vol.34, pp623-635, 2002
L. Peeters, R. de Dear, J. Hensen, and W. Dhaeseleer, Thermal
comfort in residential buildings: Comfort values and scales for building
energy simulation, Applied Energy, vol.86, pp. 772-780,May 2009
J. A. Orosa, A new modeling methodology to control HVAC systems,
Expert Systems with Applications,vol.38, pp.4505-4513, April 2011
J. Kennedy and R. Eberhart, Particle swarm optimization, Proc. IEEE
Int. Joint Conf. on Neural Network, vol.4, pp1942-1948, 1995.
F. Lin, L. Teng, J. Lin and S. Chen, Recurrent functional-link-based
fuzzy-neural-network-controlled induction generator system using
improved particle swarm optimization, IEEE Transactions on Industrial
Electronics, vol.56, pp1557-1577, May 2009
M. R. AlRashidi and M. E. El-Hawary, A survey of particle swarm
optimization applications in electric power system, IEEE Transaction
on Evolutionary Computation, vol.13, pp913-918, June 2008
B. Zhao, C. X. Guo and Y. J. Cao, A multiagent-based particle swarm
optimization approach for optimal reactive power dispatch, IEEE
Transaction on Power Systems, vol.20, pp 1070-1078, May 2005
M. Humphreys, F. Nicol, S. Roaf and O. Sykes, Standards for Thermal
Comfort Indoor air temperature standards for the 21st century,
Routledge, UK, 1995
S. H. Chen, Operations on fuzzy numbers with function principle,
Tamkang J. Manag. Sci., vol.6, pp13-25, 1985
SJ. Chen and SM. Chen, Fuzzy risk analysis based on similarity
measures of generalized fuzzy numbers, IEEE Transactions on Fuzzy
Systems, vol. 11, pp. 45-56, February 2003.
A. I. Dounis, P. Tiropanis, D. Tseles, G. Nikolaou, and G. P. Syrcos, A
Comparison of Grey Model and Fuzzy Predictive Model for Time Series,
International Journal of Engineering and Mathematical Sciences 2:3 pp.
176-181, 2006.
S.-F. Su, C.-B. Lin, Y.-T. Hsu, A high precision global prediction
approach based on local prediction approaches, IEEE SMC-C, Vol.23,
No.4, Nov. 2002, pp. 416-425.
http://www.wunderground.com

735