378 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 12, NO.

1, JANUARY 2015

Automatic Control System for Thermal Comfort Based on

Predicted Mean Vote and Energy Saving
K. L. Ku, J. S. Liaw, M. Y. Tsai, and T. S. Liu

Abstract—For human-centered automation, this study presents Index Terms—Adaptive neurofuzzy inference system, automatic
a wireless sensor network using predicted mean vote (PMV) as a air conditioning control, particle swarm algorithm, predicted
thermal comfort index around occupants in buildings. The net- mean vote (PMV), self-tuning control.
work automatically controls air conditioning by means of changing
temperature settings in air conditioners. Interior devices of air con-
ditioners thus do not have to be replaced. An adaptive neurofuzzy I. INTRODUCTION
inference system and a particle swarm algorithm are adopted for
solving a nonlinear multivariable inverse PMV model so as to de- Nowadays, most of environmental problems are closely linked to en-
termine thermal comfort temperatures. In solving inverse PMV ergy consumption. The energy consumed in buildings accounts for 40%
models, the particle swarm algorithm is more accurate than ANFIS of the total energy consumed in the entire world [1]. Moreover, air-con-
according to computational results. Based on the comfort temper- ditioning systems consume about 40%–50% of the total electricity use
ature, this study utilizes feedforward–feedback control and digital in buildings [1]. Therefore, energy control of air conditioning systems
self-tuning control, respectively, to satisfy thermal comfort. The in buildings deserves research.
control methods are validated by experimental results. Compared An air conditioning system is composed of a compressor, a con-
with conventional fixed temperature settings, the present control denser, an expansion valve or a capillary tube, and an evaporator. In
methods effectively maintain the PMV value within the range of
order to improve efficiency and maintain indoor thermal comfort, a lot
and energy is saved more than 30% in this study.
of research has been carried out to control compressors [2]–[4], [6],
Note to Practitioners—For advanced control of unitary air condi- control the opening of expansion valves [3]–[6], and control fan speeds
tioners in rooms, air conditioners may have to be retrofitted or con- of air conditioners [2], [4], [6]. For the sake of effectively controlling
nected with extra devices by wire connection, whose processes may air-conditioning units, the air conditioners may have to be dissembled
be difficult for users, and inappropriate installation may damage and the units of air conditioners may be retrofitted or connected with
original air-conditioning units. This study hence presents a nonin- extra devices by wire connection. The process may be difficult for oc-
vasive method for indoor thermal comfort with a wireless sensor cupants. In addition, temperature sensors are not always placed at de-
network. The present method facilitates hardware implementation manded spots around occupants.
without changing interior devices of the air conditioner. The wire- By contrast, this paper presents a method by means of transmitting
less sensor network measures temperature, air velocity, and hu-
the temperature commands via a wireless sensor network [7] to control
midity around occupants and further transmits temperature com-
mands for air conditioner control. Based on the measured data, air conditioner operation for occupants’ thermal comfort. The wireless
a PMV model is adopted to evaluate thermal comfort. Using an network is also utilized to obtain environment information including
inverse PMV model with feedforward–feedback control and self- the temperature, humidity, and air velocity at spots around occupants.
tuning control, respectively, this study aims to automatically main- Therefore, using the proposed control setup does not have to change
tain human thermal comfort as well as save energy. The ANFIS interior devices of existing air conditioners.
model and a particle swarm algorithm are used to solve the in- To evaluate thermal comfort, most of researches have used predicted
verse PMV model and determine the thermal comfort tempera- mean vote (PMV) model as the thermal comfort index and PMV is also
ture. Based on that temperature, feedforward–feedback control, adopted by ISO 7730 [8]–[10]. PMV takes into account six parame-
and self-tuning control are used to determine appropriate temper- ters, namely, metabolic rate, clothing insulation, air temperature, mean
ature settings in the air conditioner so as to change the cooling ca-
radiant temperature, air velocity, and humidity. According to these pa-
pacity and maintain thermal comfort. Experimental results show
that the present control method can maintain thermal comfort and rameters, PMV values represent the extent of thermal sensation. Since
saves 30% more energy than the conventional method. the temperature is the primary variable in controlling air conditioners,
an inverse PMV model is developed in this study to determine the
thermal comfort temperature dealing with desired PMV values and in-
door conditions. However, the inverse PMV model is a nonlinear and
Manuscript received August 14, 2014; revised October 20, 2014; accepted
multivariable model and it is not easy to find the analytical solution of
October 27, 2014. Date of publication November 20, 2014; date of current ver-
sion December 31, 2014. This paper was recommended for publication by As- the inverse PMV model.
sociate Editor M. Dotoli and Editor H. Ding upon evaluation of the reviewers’ Artificial intelligence strategies such as fuzzy systems, evolutionary
comments. This work was supported by the Bureau of Energy, Ministry of Eco- algorithms [11], particle swarm algorithms [12], [13], neural networks
nomic Affairs and the Green Energy and Environment Research Laboratories, [14] or the combination of the above strategies are useful for mod-
Industrial Technology Research Institute, Taiwan.
K. L. Ku and T. S. Liu are with the Department of Mechanical Engi- eling nonlinear characteristic and solving complicated problems. Par-
neering, National Chiao Tung University, Hsinchu 30010, Taiwan (e-mail: ticle swarm algorithms are inspired by a bird flock and are used to
kunlung.me97g@g2.nctu.edu.tw; tsliu@mail.nctu.edu.tw). search solutions for complex problems iteratively. Particle swarm al-
J. S. Liaw and M. Y. Tsai are with Green Energy and Environmental Lab- gorithms have features of fast searching optimal solutions and easy im-
oratories, Industrial Technology Research Institute, Hsinchu 31040, Taiwan
plementation. An adaptive neurofuzzy inference system (ANFIS) was
(e-mail: dreamyatsai@itri.org.tw; JSLiaw@itri.org.tw).
Color versions of one or more of the figures in this paper are available online presented to approximate nonlinear functions [15]. Compared with par-
at http://ieeexplore.ieee.org. ticle swarm algorithms, the ANFIS algorithm is more complex. How-
Digital Object Identifier 10.1109/TASE.2014.2366206 ever, after training ANFIS can be conducted in real time without iter-

1545-5955 © 2014 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/
redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 12, NO. 1, JANUARY 2015 379

ative computation. Particle swarm algorithms and an adaptive neuro- TABLE I

fuzzy inference system are adopted in this study to solve the inverse THERMAL SENSATION SCALE OF PMV
PMV model and obtain the desired comfort temperature, respectively.

II. INVERSE PMV MODEL

The inverse PMV model is utilized to generate the desired tem-
perature set point for controllers according to the desired PMV,
environments, and human factors. However, the inverse PMV mode is
a multivariable nonlinear model and it is difficult to find an analytical
solution. In order to solve the problem, ANFIS and particle swarm
algorithms, which belong to artificial intelligence approaches, are
adopted and their results are discussed.

A. Predicted Mean Vote in Thermal Comfort Model

The thermal comfort PMV model was proposed by Fanger [16] and
is used to predict the mean thermal comfort response. Fanger’s PMV
model establishes the relation between the thermal load on the body
and the statistical thermal sensation obtained from numerous people.
The thermal load on the body varies with personal factors and envi-
ronment factors. The personal factors consist of activity and clothing
insulation. The environment factors comprise temperature, humidity,
air velocity, and mean radiant temperature. The PMV value is calcu- Fig. 1. ANFIS architecture with two rules and two inputs.
lated by using [16]

(1)
as 0.57 clo for short-sleeved shirt with trousers [17]. In some research,
where denotes the metabolism (W/m ), and is the human load of the mean radiant temperature was assumed to equal the air temperature
the body, defined as the difference between the heat production and the due to the requirement of multiple sensors and the difficulty of mea-
heat loss to the environment [17]. The human load is computed by surement [10]. Thus, the above assumptions are used in both solving
the inverse PMV model and evaluating the PMV value in this study.

B. ANFIS for Inverse PMV

ANFIS belongs to a neurofuzzy system [15]. ANFIS can deal with
(2) nonlinear behavior and create inverse models through input/output data
[18]. Based on the input–output data, the learning capability of neural
The convection heat transfer coefficient is estimated by
networks is used to tune membership functions parameters and conse-
quent parameters in fuzzy models. The ANFIS architecture is depicted
in Fig. 1.
The ANFIS model for modeling the inverse PMV model is con-
structed in this study by using a Sugeno fuzzy model [15]. To im-
(3)
plement the ANFIS model, first of all, the training data are generated
and the surface temperature of clothing is determined by from the PMV model. Initial values of parameters in the ANFIS model
are in turn guessed. Finally, one applies an offline training technique
to tune the ANFIS parameters until meeting requirements of the error
(4) tolerance.
Fig. 2 shows the absolute thermal comfort temperature error between
where is metabolism (W/m ); is the external work (W/m ), the ANFIS output and the inverse PMV data. It can be observed that
which is equal to zero for most activity; is thermal resistance of the maximum absolute error is less than 0.015 C, which is sufficiently
clothing (clo), where (m C)/W; is the ratio of small for temperature control.
the surface area of the clothed body to that of the nude body; is the
air temperature ( C); is the mean radiant temperature ( C); is C. Particle Swarm Algorithm for Inverse PMV
the air velocity (m/s); is the partial water vapor pressure (Pa); is
the convection heat transfer coefficient (W/m ); is the surface In a particle swarm, each moving particle in a swarm is treated as a
temperature of the clothing ( C). potential solution. Each particle has memory functions, and it can mem-
PMV values are enacted based on thermal sensation scales listed in orize its current own best position and group’s best position. Like bird
Table I. It is recommended that the value of PMV should lie within the flocking, each bird adjusts its position according to its own informa-
range of to ensure indoor thermal comfort. tion and group information. The velocity of each particle in a standard
Experiments were carried out in an office in summer. The metabolic particle swarm is limited in the range of and is updated
rate and the clothing insulation of people are difficult to measure in by [13]
real time. In general, the values of the metabolism and the clothing in-
sulation are assumed as constants. The metabolic rate is assumed as
60 W/m for the office activity and the clothing insulation is assumed (5)
380 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 12, NO. 1, JANUARY 2015

Fig. 2. Absolute temperature error between ANFIS and inverse PMV data
under various PMV and relative humidity given W/m2,
m/s, and clo. Fig. 3. Absolute temperature error of the particle swarm algorithm
given W/m m/s, and
clo.

where and are the next velocity and the current velocity TABLE II
of individual , respectively; is the current position of individual COMPARISON BETWEEN ANFIS AND PARTICLE SWARM ALGORITHM
and denote the individual best position of individual
and the global best position, respectively; represents the accelera-
tion weight that pushes the particle towards represents the
acceleration weight that pushes the particle towards and
are uniformly distributed random numbers in the range is the
inertia weight. To balance global search and local search, a linearly de-
ceasing inertia weight is used and described as

(6) A. Feedforward–Feedback Control

Fig. 4 depicts the control block diagram of feedforward–feedback
where and are the maximum and minimum values of the control. The overall control action that determines the temperature set-
inertia weight, respectively; is the number of the current iteration; ting equals the sum of the feedforward control output and the feedback
is the number of maximum iteration. control output. The feedforward control simply serves to generate a
The position of each particle is limited in the range thermal comfort temperature setting. It is desired that the output value
and represented as in feedforward control is the same as the thermal comfort temperature
evaluated by an inverse PMV model. Therefore, the gain of the feed-
(7) forward controller is prescribed as 1. However, the temperature sensor
that has been embedded in air conditioners cannot reflect the temper-
In the present particle swarm algorithm, ature at spots around occupants. Therefore, the feedback controller is
, the number of particles are 40, and the maximum responsible for automatically adjusting the indoor temperature, com-
number of iterations is 30. The fitness value is the absolute error pensating the difference between the temperature measured by wire-
between the desired PMV and the PMV calculated by a candidate less sensor network around occupants and the temperature sensed by
solution. the air conditioners, and preventing the temperature around occupants
The metabolic rate is prescribed as 60 W/m and the clothing insu- from overcooling or overheating. In this study, fuzzy control and Pro-
lation 0.57 clo. The computational time is less than 1.4 s. Fig. 3 shows portional-Integral-Derivative (PID) control are respectively adopted as
temperature errors between the algorithm result and actual value. The the feedback controller.
results show that the maximum error is less than C. The
particle swarm algorithm is more accurate than ANFIS, as depicted in B. PID Controller
Table II. Particle swarm algorithms have to calculate iteratively while a
A PID controller is the most commonly used feedback controller in
well-trained ANFIS calculates the desired temperature directly without
industrial processes. Because the control algorithm is implemented by
the need of iteration.
a computer, the PID controller is written in discrete form
(8)
III. CONTROL METHODS
In order to maintain building occupants’ thermal comfort and avoid where denotes the difference between the desired temperature
wasting energy, this study utilizes an inverse PMV model with a feed- and the temperature around occupants.
forward–feedback controller and with a digital self-tuning controller,
respectively. The humidity and air velocity measured through the wire- C. Fuzzy Control
less sensor network are used in the inverse PMV model, whereas the Fuzzy control based on the fuzzy set theory [19] was developed ini-
measured temperature via the wireless sensor network is used in the tially by Mamdani [20]. Fuzzy control was also adopted to improve the
feedback controller. performance of air conditioning systems [2].
IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 12, NO. 1, JANUARY 2015 381

Fig. 4. Control block diagram of inverse PMV model with feedforward–feedback control.

In this study, input variables are the temperature error (E) and the
temperature error change (CE). The difference between the desired and
the indoor measured temperatures is E. The CE is computed by
, where e(n) is the current temperature error, is
the previous temperature error and T is the sampling time. The output
variable is the temperature change in temperature setting.

D. Digital Self-Tuning Control

A digital self-tuning controller belongs to an adaptive controller and
is suitable for time-varying system or a system whose parameters are
not completely known [21].
A regression model is employed in this study for self-tuning con- Fig. 5. Photo of a room employed for experiments. The red circle indicates the
place where a ZigBee module is located.
troller designs and the model is expressed as
(9)
where is the plant output at the sample interval, where and
is the nonmeasurable random component, is the
(15)
parameter vector described as
, and is the regression are auxiliary variables, is constant . The initial condi-
vector described as tions for recursive identification are
, and . Assume that and
where is the controller output at the in this study.
sample interval and is measurable disturbance. Takahashi’s PID controller [22] is employed in experiments of the
In addition, assume that has zero-mean value and constant digital self-tuning control
covariance.
In order to monitor slow changes in the parameters of the identified
process, the technique of adaptive directional forgetting is used. The
parameter vector can be estimated when minimizing the criterion (16)

(10) where is the reference value, is a sampling period, is the

proportional gain, is the integral time constant, and is the deriva-
where is the initial identification time, is the adaptive tive time constant. , and are tuned in real time and are deter-
directional forgetting factor, and is the mined by
prediction error. According to a recursive approach [22], the parameter
vector is computed by
(17)
(11)
where is the ultimate gain at which the output of the control loop
where is an auxiliary scalar oscillates with a constant amplitude and is the ultimate period. The
and is a rectangular covariance matrix. If , then calculation of parameters of and are based on identified model
is computed by the recurrent algorithm parameters.

IV. EXPERIMENTAL SETUP

(12) The room size of the office selected for executing experiments is
where 6 m in length, 5.8 m in width, and 2.8 m in height. The photo of the
(13) room is shown in Fig. 5. Furthermore, there are three people in the
room when performing experiments. A unitary air conditioner with a
If , then . The value of adaptive nominal cooling capacity of 7.3 kW is employed for experiments.
forgetting factor is updated by the relation Compared with wire-based measurement and control systems, wire-
less systems [23] have the advantage of easy installation, convenience
of relocation, and expansion for equipment. Wireless technologies ap-
plied in building automation systems include ZigBee (IEEE 802.15.4
(14)
protocol), Wi-Fi (IEEE 802.11 protocol standard), Bluetooth, etc., we
382 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 12, NO. 1, JANUARY 2015

Fig. 6. Signal transmission among devices in the wireless control network.

use a ZigBee module including an infrared emitter for air conditioner

remote control and environment sensors that can measure temperature,
humidity, air velocity, illuminance, and CO concentration. In addition, Fig. 7. Measured PMV responses of the conventional method, the inverse PMV
the real-time environment information can be displayed on the monitor model with feedforward-fuzzy feedback control and self-tuning control.
in the ZigBee module. However, only the temperature, air velocity, and
humidity functions are employed for thermal comfort control in this TABLE III
study. The ZigBee module is placed on the desk around occupants and COMPARISON OF MEASURED PMV PERFORMANCES IN
FOUR CONTROL METHODS
located at the red circle shown in Fig. 5.
Fig. 6 depicts signal transmission among devices in the overall wire-
less network. And it has two coordinators: one is USB-type coordinator
and it is connected to the ZigBee measurement and control module, and
the other is connected to other devices including wattmeter of the air
conditioner, lightings, and printer, and other devices.

V. EXPERIMENTAL RESULTS
Experiments were performed for four control methods in Taiwan,
where the climate is a hot and subtropical humid. The first control
method is the conventional method, i.e., fixed temperature setting at
26 C. The other three control methods belong to the inverse PMV
model with the feedforward-fuzzy feedback control, with feedfor-
ward-PID feedback control, and with digital self-tuning control. The
air conditioner is operated to track the thermal comfort temperature by
the last three controllers. The inverse PMV model computes thermal
comfort temperature in real time based on the desired PMV, measured thermal comfort and yields smaller variation of PMV than the conven-
air velocity, and humidity. In order to ensure PMV in the range of tional method.
and reduce energy consumption, this study prescribes the Among these three controllers, the feedforward-fuzzy feedback
PMV input in the inverse PMV model depicted in Fig. 4 as 0.25 in the controller has the largest percentage of periods lying within PMV
summer, during which experiments were conducted for 4.5 hours. By and the smallest variation of PMV than the other con-
contrast, in the winter the PMV input of is suggested due to trollers. Comparing between the feedforward-PID feedback controller
energy saving consideration. and the digital self-tuning controller. Table III shows that the dig-
Performances of the four methods are compared based on PMV re- ital self-tuning controller performs better than the feedforward-PID
sponse curves shown in Fig. 7. The PMV value of the conventional feedback controller in terms of the percentage of the period lying
method changes more severely than the other three controllers. Ac- within PMV and the PMV variation. Table III also shows
cording to Fig. 7, the PMV values of the three nonconventional con- that the effective tuning of the self-tuning control parameters results
trollers maintain between 0 and 0.5. The three perform better than the in better control performance than the feedforward-PID feedback
conventional method because the inverse PMV model can real time controller. Furthermore, among these controllers, the average PMV of
generate proper comfort temperatures, which are in turn continuously the digital self-tuning controller is closest to the PMV input than the
tracked by each of three controllers. Furthermore, according to the other controllers.
measured indoor temperature around occupants and the thermal com- Table IV compares the energy consumptions among the four
fort temperature, the three nonconventional controllers appropriately methods. The energy saving is calculated according to
change the temperature setting in the air conditioner, which is equiva-
lent to adjusting cooling capacities at any time. % (18)
Table III compares performances among control methods. The per-
centage of the period lying within PMV is defined as the where is the energy consumption of the conventional method,
ratio of periods during which the PMV value lies within and is the energy consumption of the ANFIS with the three con-
to the total time from the time at which the PMV first enters within trol methods. According to Table IV, the inverse PMV model with
to the end time. The percentages of periods lying within feedforward-fuzzy feedforward-PID or with digital self-tuning con-
PMV of the inverse PMV model with the three noncon- trol indeed outperform the conventional method by 34.7%, 37.3%, and
ventional controllers are larger than the conventional one. And the stan- 32.9%, respectively, in energy saving. The inverse PMV model with
dard deviations of the three are smaller than the conventional method. the feedforward-fuzzy feedback control saves the most energy among
The inverse PMV model with these three controllers can better satisfy the three methods.
IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 12, NO. 1, JANUARY 2015 383

TABLE IV [4] N. Jain, B. Li, M. Keir, B. Hencey, and A. Alleyne, “Decentralized

COMPARISON OF MEASURED ENERGY CONSUMPTION AND feedback structures of a vapor compression cycle system,” IEEE Trans.
ADVANTAGES AMONG FOUR CONTROL METHODS Control Syst. Technol., vol. 18, no. 1, pp. 185–193, Jan. 2010.
[5] D. P. Finn and C. J. Doyle, “Control and optimization issues associated
with algorithm-controlled refrigerant throttling devices,” ASHRAE
Trans., vol. 106, no. 1, pp. 524–533, 2000.
[6] T. J. Yeh, Y. J. Chen, W. Y. Hwang, and J. L. Lin, “Incorporating fan
control into air-conditioning systems to improve energy efficiency and
transient response,” Appl. Therm. Eng., vol. 29, no. 10, pp. 1955–1964,
2009.
[7] J. Y. H. Wang, Y. K. Chuah, S. W. Chou, and T. H. Lo, “Non-invasive
Zigbee wireless controller for air conditioner energy saving,” in Proc.
IEEE 7th Conf. WiCOM, Wuhan, 2011, pp. 1–4.
[8] J. Liang and R. Du, “Design of intelligent comfort control system with
human learning and minimum power control strategies,” Energy Con-
vers. Manage., vol. 49, no. 4, pp. 517–528, 2008.
[9] Moderate Thermal Environments—Determination of the PMV and
PPD Indices and Specification of the Conditions for Thermal Comfort,
International Standard ISO 7730, 1994.
[10] Y. Farzaneh and A. A. Tootoonchi, “Controlling automobile thermal
comfort using optimized fuzzy controller,” Appl. Therm. Eng., vol. 28,
no. 14, pp. 1906–1917, 2008.
VI. CONCLUSION [11] D. Niu, Y. Sun, and F. Wang, “Optimization of advertising budget al-
location over time based on LS-SVMR and DE,” IEEE Trans. Autom.
Experiments have been carried out by using four control methods. Sci. Eng., to be published.
As shown in Fig. 7, PMV response curves of every controller fluctu- [12] W. Kong, T. Chai, J. Ding, and S. Yang, “Multifurnace optimization in
ates due to 1 C increment of air-conditioner temperature commands. electric smelting plants via load scheduling and control,” IEEE Trans.
Autom. Sci. Eng., vol. 11, no. 3, pp. 850–862, Jul. 2014.
Therefore, it remains to develop methods and devices to maintain PMV [13] X. Q. Zuo, G. X. Zhang, and W. Tan, “Self-adaptive learning PSO-
near 0 and smooth responses while saving energy. In addition, since the based deadline constrained task scheduling for hybrid IaaS cloud,”
comfort range of PMV may vary with different people’s feeling, it may IEEE Trans. Autom. Sci. Eng., vol. 11, no. 2, pp. 564–573, Apr. 2014.
[14] T. Songuppakarn, W. Wongsuwan, and W. San-um, “Artificial neural
be required to modify the comfort range of PMV according to the ques- networks based prediction for thermal comfort in an academic class-
tionnaire of occupants’ preference or develop other thermal comfort in- room,” in Proc. ICUE, 2014, pp. 1–8.
dices for occupants. In this study, the values of the metabolic rates and [15] J. S. R. Jang, “ANFIS: Adaptive-network-based fuzzy inference
the clothing insulation are assumed as constants and are estimated from system,” IEEE Trans. Syst, Man, Cybern., vol. 23, no. 3, pp. 665–685,
May/Jun. 1993.
tables in [17]. For accurate estimation, the clothing insulation can be [16] P. O. Fanger, Thermal Comfort: Analysis and Applications in Environ-
determined by measurement on heated mannequins, and the metabolic mental Engineering. New York, NY, USA: McGraw-Hill, 1970.
rates can be estimated from measuring CO and O in a person’s ex- [17] ASHRAE Handbook of Fundamentals. Atlanta, GA, USA: ASHRAE,
2010.
pired air [17]. Moreover, human activity changes with time. However, [18] A. G. Martin and R. E. H. Guerra, “Internal model control based on a
estimating metabolic rates and clothing insulation is not trivial. There- neurofuzzy system for network applications: A case study on the high-
fore, it is desired in future work to devise wearable or non-contact sen- performance drilling process,” IEEE Trans. Autom. Sci. Eng., vol. 6,
sors to measure the values of metabolic rates and clothing insulation no. 2, pp. 367–372, Apr. 2009.
[19] L. A. Zadeh, “Fuzzy sets,” Inform. Control, vol. 8, no. 3, pp. 338–353,
and improve the human factor measurement process. 1965.
[20] E. H. Mamdani, “Application of fuzzy algorithms for control of simple
dynamic plant,” Proc. IEEE Inst. Elect. Eng., vol. 121, no. 12, pp.
REFERENCES 1585–1588, 1974.
[1] B. Sun, P. B. Luh, Q. S. Jia, Z. Jiang, F. Wang, and C. Song, “Building [21] V. Bobál, J. Böhm, J. Fessl, and J. Macháček, Practical Aspects of Self-
energy management: Integrated control of active and passive heating, tuning Controllers: Algorithms, Implementation and Applications.
cooling, lighting, shading, and ventilation systems,” IEEE Trans. London, U.K.: Springer-Verlag, 2005, 2005.
Autom. Sci. Eng., vol. 10, no. 3, pp. 588–602, Jul. 2013. [22] V. Bobál, J. Böhm, and R. Prokop, “Practical aspects of self-tuning
[2] C. B. Chiou, C. H. Chiou, C. M. Chu, and S. L. Lin, “The application of controllers,” Int. J. Adapt. Control Signal Process., vol. 13, pp.
fuzzy control on energy saving for multi-unit room air-conditioners,” 671–690, 1999.
Appl. Therm. Eng., vol. 29, no. 2, pp. 310–316, 2009. [23] J. W. Jeong, J. Firrantello, J. D. Freihaut, W. P. Bahnfleth, and A.
[3] X. D. He, S. Liu, H. H. Asada, and H. Itoh, “Multivariable control Musser, “Feasibility of wireless measurements for semi-empirical
of vapor compression systems,” HVAC&R Res., vol. 4, no. 3, pp. multizone airflow model tuning,” Build. Environ., vol. 43, no. 9, pp.
205–230, 1998. 1507–1520, 2008.