Building Serv. Eng. Res. Technol. 22,4 (2001) pp.

237–253

Thermal comfort based fuzzy logic controller

MM Gouda BSc MSca, S Danaher BSc PhD CPhys CEng MIEEa and CP Underwood BSc PhD CEng
MCIBSE MASHRAEb
aSchool of Engineering, University of Northumbria, Newcastle upon Tyne, UK
bSchool of the Built Environment and Sustainable Cities Research Institute, University of Northumbria,
Newcastle upon Tyne, UK

Most heating, ventilation and air conditioning (HVAC) control systems are considered as
temperature control problems. In this work, the predicted mean vote (PMV) is used to
control the indoor temperature of a space by setting it at a point where the PMV index
becomes zero and the predicted percentage of persons dissatisfied (PPD) achieves a
maximum threshold of 5%. This is achieved through the use of a fuzzy logic controller
that takes into account a range of human comfort criteria in the formulation of the
control action that should be applied to the heating system to bring the space to comfort
conditions. The resulting controller is free of the set up and tuning problems that hinder
conventional HVAC controllers. Simulation results show that the proposed control strat-
egy makes it possible to maximize the indoor thermal comfort and, correspondingly, a
reduction in energy use of 20% was obtained for a typical 7-day winter period when
compared with conventional control.

List of symbols Met human metabolic rate (W/m2)

pv relative air velocity (m/s)
To outdoor air temperature (°C) fcl ratio of body’s surface area when fully
Ti internal air temperature (°C) clothed to body’s surface area when nude
Tj surface temperature of the jth construc- Icl thermal resistance of the clothing (Clo)
tion element (°C) hc convection heat transfer coefficient
Tmrt mean radiant temperature (°C) (W/m2K)
Tcl surface temperature of clothing (°C) M mass of air in a room (kg)
Aj surface area of the jth construction mo input ventilation mass flow rate (kg/s)
element (m2) Ql latent heat gain (kJ)
fi internal relative humidity (%) h Specific enthalpy of water vapour
fext outdoor air relative humidity (%) (kJ/kg)
S percentage saturation C Specific heat capacity of air (kJ/kgK)
g moisture content of air (kg/kg)
gss moisture content of air at saturation
(kg/kg) 1 Introduction and review
Hr internal air enthalpy (kJ/kg)
Ho outdoor air enthalpy (kJ/kg) Over the years, a large number of thermal comfort
Pwsat partial pressure of saturated water- indices have been established for the analysis of
vapour at a given temperature (N/m2) indoor climates and the design of heating, venti-
a,b constants lation and air conditioning (HVAC) systems.1–4
Several have been used to assess the extent to
which an existing room climate achieves satisfac-
Address for correspondence: CP Underwood, School of the Built
Environment, University of Northumbria, Newcastle upon Tyne tory comfort conditions for occupants. A thermal
NE1 8ST, UK. E-mail: [email protected] comfort index, called effective temperature (ET),
© The Chartered Institution of Building Services Engineers 2001 0143–6244(01)BT021OA

Downloaded from bse.sagepub.com at NANYANG TECH UNIV LIBRARY on April 26, 2015
238 Fuzzy logic controller

has been proposed as a combination of indoor plant. However, they do require information on
temperature (Ti) and indoor relative humidity plant dynamics during commissioning since the
(fi). This concept was further developed and proportional, integral and derivative gain
adopted as a thermal comfort index by constants of this controller require to be deter-
ASHRAE5 and has been widely adopted as a mined either by manual tuning or by online
design and comfort performance criterion for tuning in some way. Since these constants
decades. depend on system parameters in most cases, re-
Fanger1 published a general comfort equation tuning is necessary when the parameters them-
which makes it possible, for specified activity selves change (e.g., due to changes in plant
and clothing levels, to calculate all combinations operating point, operating mode, malfunction or
of the environmental variables (air temperature, modifications to the plant itself).
air humidity, mean radiant temperature, and rela- One alternative to this, which requires no
tive air velocity) which will create optimal ther- knowledge of plant dynamics, is fuzzy logic
mal comfort. In order to derive the comfort control (FLC).11 This, coupled with the vague-
equation, Fanger supposed that for long expo- ness and subjectivity of thermal comfort in prac-
sures to a constant thermal environment with a tice, make fuzzy logic well suited for the
constant metabolic rate, a heat balance can be evaluation and control of thermal sensation as a
established for the human body such that bodily fuzzy concept in which the comfort range can be
heat production is equal to heat dissipation. evaluated as a fuzzy range rather than an isolated
Based on these assumptions, Fanger developed a comfort variable.
thermal sensation index to predict the mean ther- The PMV-based fuzzy logic controller investi-
mal sensation vote on a standard scale for a large gated in this work starts with the evaluation of
group of persons depending on the four thermal the predicted mean vote level and compares this
environmental variables, together with a defined with the required comfort range in order to arrive
activity level and ‘clo-value’ based on clothing at a linguistic definition of the comfort sensation.
worn by the occupants. He defined this as a The controller then adjusts the air temperature set
‘predicted mean vote’ (PMV) and, concomi- point in order to satisfy the required comfort
tantly, a ‘percentage of persons dissatisfied level given the prevailing values of other comfort
(PPD) was also defined. ISO Standard 77306 has variables contributing to the comfort sensation.
adopted Fanger’s proposals based on a PMV of 0 The objectives of this work are thus:
with a tolerance of ±0.5.
• To develop a model sufficient for investigating
The role of HVAC services is to provide these
the comfort sensation control of the heating
standards of comfort at minimum energy use and
system in a building space.
it is the HVAC control system that has the crucial
• To develop a control strategy which responds
influence over this. Previous work has shown
to the essentially subjective basis of comfort
that it is possible to reach these objectives if
sensation.
HVAC control strategies are based on a thermal
• To compare the developed control strategy
sensation index instead of air temperature
with a conventional method of building space
alone1,7,8 and the advantages of such a strategy
heating control.
have been reported in terms of improved thermal
comfort performance and energy saving.9,10
Presently, the proportional-integral-derivative 2 System model
controller continues to satisfy most non-domestic
HVAC applications. One of the reasons for their The dynamics of a building space depend on
popularity is that they require little a priori infor- external microclimate variables, building
mation about the plant and do not necessarily construction, user influences (e.g., adventitious
require detailed modelling information about the internal heat generation) and the imposed HVAC

Downloaded from bse.sagepub.com at NANYANG TECH UNIV LIBRARY on April 26, 2015
 MM Gouda, S Danaher and CP Underwood 239

plant. Thermal response modelling of building extent. This is particularly evident when the
spaces has received widespread attention over the following prevail (either individually or in
past 30 years or so, however, the treatment of this combination):
problem over the relatively short time periods of
• large spaces;
importance in control system design and synthe-
• irregularly shaped spaces;
sis has received very limited attention. Particular
• where there is a high degree of glazing;
to this problem is a sufficiently accurate descrip-
• where bare heating surfaces form the main
tion of the building envelope that is computation-
method of space heating.
ally efficient at those time intervals necessary for
fully dynamic HVAC plant participation. Accounting for these conditions in the assess-
A model which addresses the main features of ment of Tmrt is feasible in theory but not easy in
this problem is described elsewhere,12,13 and a practical situation in which a single point
adapted for the specific purpose of this work in the measurement is generally used to inform control.
following. In principle, three adaptations (besides Part of the problem is concerned with the point of
internal air temperature) were needed in order to measurement since the occupant sensation of
prosecute a full thermal comfort sensation analy- comfort is spatially dependent. In small spaces
sis but only two of these were incorporated. The this is not significant and in a large space it can
three adaptations were mean radiant temperature; be dealt with by dividing the space into a number
relative humidity; and mean relative air velocity. of rectilinear sub-spaces with a point of measure-
Only the first two were incorporated in the model ment in each. A further possibility for those
developed previously since it was assumed that spaces exhibiting one or more of the features
for the heated and naturally ventilated case of mentioned above is to implement a model-
interest in this work, that mean relative air veloc- assisted strategy in which a thermal model of the
ity (besides being difficult to evaluate dynami- space is used to offset measurements made at a
cally) would be insignificant in most applications. single point. None of these avenues have been
explored in the present work and form the basis
2.1 Mean radiant temperature of further work in the field.
The mean radiant temperature has a significant
influence on the body’s rate of heat loss and thus 2.2 Relative humidity
its comfort state as defined by the Fanger Relative humidity in the space is calculated
PMV.1,14 Calculation of the mean radiant temper- using standard psychrometric properties of
ature is complicated by the non-uniform ‘view’ air.15,16 Details of the algorithm used can be
that the body has of the various surfaces making found in Appendix A. The uniformity of this
up a room space. Thus an approximation has been parameter within the space is again an issue (see
used in the present work, and the mean radiant above concerning Tmrt) however in most cases
temperature has been calculated by the following the air moisture content (on which the relative
equation based on an area-weighted mean in humidity is dependent together with prevailing
which the index nc refers to the number of air temperature) will be well mixed unless the
instruction elements involved: space is very large. In any case the relative
nc
humidity is less influential on overall comfort
å Aj Tj sensation than is the case with temperature vari-
ables. Again, non-uniformity in relative humidity
Tmrt = 1
nc (1) is not considered in the present work.
å Aj
1
2.3 Input data
In practice, most building spaces experience Input data for the adapted system model were
asymmetry in the radiant field at least to some based on a space in a campus building at the

Downloaded from bse.sagepub.com at NANYANG TECH UNIV LIBRARY on April 26, 2015
240 Fuzzy logic controller

University of Northumbria. The building exhibits when ‘cool’ or ‘cold’. The PPD is estimated to be
relatively high thermal capacity on account of its 5% when the PMV is zero, and the target indoor
traditional construction and the heating system temperature is set with respect to this point.
comprises hot water natural convectors with
local feedback control. 4 Conventional comfort-based control

3 Thermal comfort indices In order to identify a reference performance for a

comfort-based control strategy, a conventional
The thermal comfort sensation is expressed using approach was first developed based on a
Fanger’s predicted mean vote (PMV) and proportional-plus-integral-plus-derivative (PID)
percentage persons dissatisfied (PPD)1 in which controller, a method commonly used in space
the environmental factors of influence are room heating control.17,18 The resulting controller and
air temperature, relative humidity, mean radiant feedback path were added to the adapted system
temperature, and air velocity. In the present model as summarized in Figure 1. The controlled
work, a constant mean was used for the air veloc- variable (PMV) is calculated at each time step
ity. In addition, there are human factors associ- from prevailing values of room air temperature,
ated with activity level, and the thermal relative humidity and mean radiant temperature
resistance of clothing. The relationship between with the room air velocity, activity and clothing
PMV, PPD and the thermal sensation of the occu- levels treated as constants. This is compared with
pant is summarized in Table 11 based on a a reference value of PMV and the resulting error
neutrality at zero PMV (positive when the ther- is applied to the controller. External temperature,
mal sensation is ‘warm’ or ‘hot’ and negative solar radiation and casual heat gains act as distur-
bances, as shown in Figure 1. The output of the
PID controller is given by:
Table 1 Relationship between PMV, PPD and thermal sensa- t
de(t )
tion1 u(t ) = kp e(t ) + ki ò e(t ) + kd (2)
0
dt
PMV Thermal sensation PPD (%)

+3 Hot 100 Where e(t) is the error, u(t) is the control

+2 Warm 75 signal and kp, ki, kd are proportional, integral and
+1 Slightly warm 25
0 Neutral 5
derivative gain constants, respectively.
–1 Slightly cool 25 The initial parameters of the controller were
–2 Cool 75 based on those actually in use in the building
–3 Cold 100
space to which the model was applied (see

Figure 1 Closed loop system

Downloaded from bse.sagepub.com at NANYANG TECH UNIV LIBRARY on April 26, 2015
 MM Gouda, S Danaher and CP Underwood 241

Section 2.3), these values being 0.1K–1, the planes of minimum normalized error to initiate
0.01K–1s–1, and 0.5sK–1 (representing kp, ki, kd the tuning algorithm, tuned controller parameters
respectively). This produced a reasonable of 1.527K–1, 0.005K–1s–1, and 0.050sK–1 respec-
response but with overshoot (Figure 5), leading tively were obtained for kp, ki, kd. In this way, the
to sub-optimal thermal comfort and unnecessary initial mapping resulted in substantial reductions
use of energy. in computational effort needed by the tuning algo-
A tuning algorithm was adopted to adjust the rithm. The response of the system under tuned PID
controller parameters in order to eliminate this controller is shown in Figure 5 showing an
overshoot. The MATLAB Non-linear Control improvement over the existing case.
Design (NCD) Blockset was used19 – a gradient-
based optimization designed to minimize a cost 5 Fuzzy logic control (FLC)
function (i.e., a weighted maximum constraint
violation of the constrained (control) variable) Under well-tuned PID control, performance is
with reference to perturbed values of the tunable excellent within the narrow operating range
variables (controller parameters). To give initial within which the plant was tuned as is evident
estimates of the controller parameters, an error above. Once the plant operating region changes
mapping method was used based on the initial significantly (e.g., as a result of a change in
parameters given above as shown in Figures 2, 3, season), then the need for retuning becomes
and 4. Using the initial mapped values from within evident. Also, it is not possible to adequately

Figure 2 Normalized error versus kp, ki

Downloaded from bse.sagepub.com at NANYANG TECH UNIV LIBRARY on April 26, 2015
242 Fuzzy logic controller

Figure 3 Normalized error versus kd, ki

Figure 4 Normalized error versus kd, kp

Downloaded from bse.sagepub.com at NANYANG TECH UNIV LIBRARY on April 26, 2015
 MM Gouda, S Danaher and CP Underwood 243

Figure 5 Closed loop response with PID control

generalize the required parameter specification for lems will be described. More fundamental infor-
different applications – for example a PID mation can be found in the literature.20–23 The
controller optimized for the control of heating in a static fuzzy controller consists of four main func-
higher thermal capacity space will be sub-optimal tional blocks (Figure 6): fuzzification interface;
and possibly even unstable in a low thermal capac- fuzzy control rules; inference engine, and
ity space. Similar difficulties arise with the wide defuzzification interface.
choice of heating systems that can be applied. In
addition, in the case of comfort control, a desired 5.1 Fuzzification interface
comfort sensation will vary with both application The fuzzification interface consists of the
and season and is, in any case, subjective (Table following operations:
1). One solution is a controller that can respond to
this essentially subjective problem with experien- (1) Performs a scale mapping that transfers the
tial information about plant response and user input variable ranges into a corresponding
requirements. A fuzzy logic controller has the universe of discourse (quantization/normal-
potential to meet these needs. ization).
In this section the basic structure of the fuzzy (2) Performs the fuzzification strategy that
logic controller applied to comfort control prob- converts crisp input data into suitable

Downloaded from bse.sagepub.com at NANYANG TECH UNIV LIBRARY on April 26, 2015
244 Fuzzy logic controller

Figure 6 Structure of a fuzzy logic controller

linguistic variables, which may be viewed as depends on the problem to be dealt with and the
labels of fuzzy sets. choice of the number of fuzzy sets and how those
fuzzy sets are divided over the universe of
The fuzzification strategy converts the crisp discourse depends generally on how the
input data into fuzzy sets (linguistic variables) controller output should be related to the
such as, cold (COLD), cool (COOL), slightly controller input. For example, designing a fuzzy
cool (SCOOL), neutral (NEUT), slightly warm controller for controlling a non-linear process
(SWARM), warm (WARM), and hot (HOT). The requires knowledge of the non-linearity of the
fuzzification action consists of a set of analogue process. There is no standard design scheme that
membership functions, describing the input can be employed to choose the number and posi-
linguistic terms. The membership function can be tion of the fuzzy sets, and too few people realize
of a variety of shapes (e.g., triangle, trapezoid, that this is a problem.
etc.). The overlapping of fuzzy sets together with
When building a fuzzy controller, one of the fuzzy inference and defuzzification result in
first questions which arises, after having chosen interpolation. If the membership functions are
the inputs, is how many fuzzy sets will be needed convex and normal and the sets are a fuzzy parti-
and how the fuzzy sets should be divided on the tion, then the interpolation depends only on the
universe of discourse of the inputs. More or less nearest surrounding characteristic set of points
standard types of fuzzy sets on a universe of and each characteristic point is uniquely defined
discourse of controllers are shown in Figure 7(a) by a fuzzy rule. This is because there are no more
and (b). than two overlapping membership functions at
The choice of membership functions used any point on the universe of discourse.

Downloaded from bse.sagepub.com at NANYANG TECH UNIV LIBRARY on April 26, 2015
 MM Gouda, S Danaher and CP Underwood 245

a)

b)

Figure 7 Types of membership functions: (a) linear and (b) logarithmic

Downloaded from bse.sagepub.com at NANYANG TECH UNIV LIBRARY on April 26, 2015
246 Fuzzy logic controller

5.2 Fuzzy control rules be used by the plant in order to create indoor
The dynamic behaviour of a fuzzy system is thermal comfort. The controller, making it easy
characterized by a set of imprecise conditional to apply and generic to a wide range of heating
statements, which form a set of decision rules. The control problems uses a linguistic description of
process can be expressed linguistically as a set of the thermal comfort sensation. Overlapping
linguistic decision rules of the form: If (Conditions triangular membership functions were used for
are satisfied) Then (Action can be inferred). input (fuzzification) and output (defuzzification)
There are four ways to derive fuzzy control of the fuzzy system.
rules:25,26 The input membership functions were defined
by assigning seven fuzzy input sets to the vari-
• From experiential knowledge of the plant
ables; (COLD, COOL, SCOOL, NEUT, SWARM,
behaviour and control problem.
WARM, and HOT), as shown in Figure 8.
• By modelling or observing a human operator’s
Seven fuzzy sets were ascribed to the output
manual interpretation of the problem.
variables; (FC, CL, SCL, MDL, SOP, OP, FO) to
• From a model of the plant behaviour and
form the output membership functions as shown
control problem.
in Figure 9.
• Self-learning and adaptation by the controller
According to the number of the fuzzy sets of
(self-organization).
the input and the output, seven fuzzy rules may
be defined as follows:
5.3 Inference
There are in general four methods of fuzzy R1: If PMV is ‘COLD’ Then Vp is ‘FO’
reasoning.24 Mamdani’s minimum operation R2: If PMV is ‘COOL’ Then Vp is ‘OP’
method is the most common, involving two func- R3: If PMV is ‘SCOOL’ Then Vp is ‘SOP’
tions: R4: If PMV is ‘NEUT’ Then Vp is ‘MDL’
R5: If PMV is ‘SWARM’ Then Vp is ‘SCL’
1) Determine for any fuzzy controller input
R6: If PMV is ‘WARM’ Then Vp is ‘CL’
which rules are applicable.
R7: If PMV is ‘HOT’ Then Vp is ‘FC’
2) Determine the fuzzy control action by using
fuzzy reasoning. The relationship between the input and the
output of the controller according to these fuzzy
5.4 Defuzzification strategies rules is shown in Figure 10.
At the output of inference the process is a Using Mamdani’s minimum operator method
fuzzy set (i.e., for example, Figure 7). A non- for inference, the control action is a fuzzy set,
fuzzy control signal (i.e., a crisp value) can be which requires a defuzzification strategy to
established through a defuzzification. There are obtain a crisp control signal. The COG method
several methods for defuzzifying to a crisp was used to convert from fuzzy values to crisp
value25 of which the centre of gravity method values, forming the actual control signal, which
(COG) is the most commonly used. This method can then be applied to the heating valve.
is based on taking the aggregate of the fuzzy The overall system (building, heating system,
outputs from each rule, weighted by their grades and outdoor climate files) with the PMV based
of fuzzy input set membership. fuzzy logic controller and, for comparison, the
PID-based comfort controller is shown in Figure
6 A PMV-based fuzzy logic controller 11. A switch has been added to the model, before
the non-linear valve, to select between PID and
The PMV-based fuzzy logic controller proposed fuzzy logic control actions. Two different
here evaluates the predicted mean vote (PMV) outdoor climates have been used; one in
level and, if this level is out of the comfort range, February (a winter month) and the second in
provides the air temperature set point that should April (a spring month).

Downloaded from bse.sagepub.com at NANYANG TECH UNIV LIBRARY on April 26, 2015
 MM Gouda, S Danaher and CP Underwood 247

Figure 8 Membership functions of the fuzzy controller input

The calculation of the predicted mean vote typical winter conditions and the energy
(PMV) and the predicted percentage of persons demands in both cases were integrated to reveal
dissatisfied (PPD) is determined using a an energy saving of approximately 20% due to
MATLAB function based on three inputs from the the PMV-based FLC when compared with
simulation model; the internal air temperature, the conventional tuned PID control.
mean radiant temperature, and the relative humid- In order to test the general applicability of the
ity of the internal air. Because the activity and two control systems, the building construction
clothing levels are application-dependent, these data were changed to those representing a very
values were entered as constants. low thermal capacity structure while keeping the
The resulting PMV based fuzzy logic overall thermal transmittance of each element
controller was applied to the building space and constant and the controller specifications
its heating system. Figure 12, shows the system remained unchanged. This would have the effect
response under tuned PID and PMV-based FLC. of making the system much more responsive thus
As can be seen, the performance of the PMV challenging the PID controller originally tuned
based FLC is superior to PID controller in terms for a high thermal capacity application. Results,
of reference point tracking. Results were shown in Figure 13, reveal that the fuzzy
extended over a 7-day cycle of space heating in controller maintains excellent tracking of the

Downloaded from bse.sagepub.com at NANYANG TECH UNIV LIBRARY on April 26, 2015
248 Fuzzy logic controller

Figure 9 Membership functions of the fuzzy controller output

Figure 10 Input/output relationships for the fuzzy controller

Downloaded from bse.sagepub.com at NANYANG TECH UNIV LIBRARY on April 26, 2015
 MM Gouda, S Danaher and CP Underwood 249

Figure 11 Overall model of the building, heating system and controllers

Figure 12 Comparison of tuned PID and PMC-based FLC

Downloaded from bse.sagepub.com at NANYANG TECH UNIV LIBRARY on April 26, 2015
250 Fuzzy logic controller

Figure 13 Low thermal capacity building with PID control and PMC-based FLC

reference condition whereas the tuned PID gives PID control of this problem are identified and an
a reasonable response but with significant over- alternative strategy based on a fuzzy logic
shoot. Thus the robustness characteristic of the controller (FLC) is developed. There is a natural
PMV-based FLC is superior to that of the PID appeal in this in that the essentially subjective
controller. In practice, the PID controller would comfort control problem can be easily mapped
either have to be tuned for each individual appli- onto a universe of discourse of input fuzzy sets and
cation, or tuned for an expected least-damped the associated inference which ultimately forms a
case (in which case it would behave sub-opti- required control action can be carried out qualita-
mally in all other applications) whereas the tively. The resulting FLC is compared with a PID
PMV-based FLC would not require this. controller tuned to give optimal performance for a
given case, consisting of a building space with high
7 Conclusions thermal capacity. A further comparison is then
made using the same controller specifications but
In this work, a control strategy has been developed with a low thermal capacity building space. Results
for the space heating of a building space in which show that the FLC gives better control tracking and
the comfort sensation as quantified by predicted robustness than the PID controller for both appli-
mean vote forms the control variable. Difficulties cations. This work has addressed the extremes of
associated with the lack of generality inherent in building space thermal capacity in arriving at these

Downloaded from bse.sagepub.com at NANYANG TECH UNIV LIBRARY on April 26, 2015
 MM Gouda, S Danaher and CP Underwood 251

conclusions and further work needs to be done on 7 Auliciems A. Thermobile controls for human
the applicability of the results to a wider range of comfort. The Heating and Ventilation Engineer
construction types, heating system types (and 1984; 31–3.
modes) and microclimate conditions. It is likely 8 Fountain M, Brager G, Arens E, Bauman F,
that some degree of self-adaptation will therefore Benton C. Comfort control for short-term occu-
pancy. Energy and Buildings 1994; 21: 1–13.
be necessary and a neuro-fuzzy approach is 9 Culp CH, Rhodes ML, Krafthefer BC, Listvan
presently being actively explored. Further work is M. Silicon infrared sensors for thermal comfort
also needed on the essentially spatially-variable and control. ASHRAE Journal 1993; 38–42.
nature of certain variables which contribute to 10 MacArthur JW. Humidity and predicted mean
comfort sensation (Sections 2.1, 2.2). One area vote based comfort control. ASHRAE Trans.
meriting further work is the use of model-assisted 1986; 1: 5–17.
compensation in which a thermal model is used to 11 King PJ, Mamdani EH. The application of fuzzy
predict the spatial distribution of the variables of control systems to industrial processes.
interest with reference to a single point of measure- Automatica 1977; 13: 235–42.
ment. From this, the measurement can subse- 12 Gouda MM, Danaher S, Underwood CP.
quently be adjusted to account of non-uniformity Modelling the heating of a building space using
Matlab and Simulink. Proceedings of IMACS
within the space. Symposium on Mathematical Modelling, Vienna,
2000; 1: 91–4.
Acknowledgements 13 Gouda MM, Danaher S, Underwood CP. Low-
order model for the simulation of a building and
The authors would like to thank the Northumbria its heating system. Building Serv. Eng. Res.
Photovoltaics Application Centre at the Technol. 2000; 21: 203–12.
University of Northumbria and Meteorological 14 Dounis AI, Santamouris MJ, Lefas CC,
Office (Academic Research Section) in London, Manolakis DE. Thermal comfort degradation by
for supplying the weather data used for this work. a visual comfort fuzzy reasoning machine under
Also, the authors express their gratitude to the natural ventilation. Applied Energy 1994; 48:
Egyptian government for financial support. 115–30.
15 CIBSE Guide part C, section C1.
References 16 Psychrometrics: theory and practice. Georgia:
American Society of Heating, Refrigerating, and
1 Fanger PO. Thermal comfort, analysis and appli- Air-Conditioning Engineers, Inc, 1996.
cation in environmental engineering. McGraw- 17 ASHRAE, Systems handbook. Georgia:
Hill, 1972. ASHRAE Inc, 1984.
2 Gagge AP, Fobelets AP, Berglund LG. A standard 18 Haines RW. HVAC systems design handbook.
predictive index of human response to the ther- USA: TAB Books, 1988.
mal environment. ASHRAE Trans 1986; 92: 19 MathWorks. Matlab user manual. MA: Natick,
709–31. 1993.
3 Sherman M. A simplified model of thermal 20 Lee CC. Fuzzy logic in control systems: fuzzy
comfort. Energy and Buildings 1985; 8: 37–50. logic controller part 1. IEEE Trans. System Man.
4 Hamdi M, Lachiver G, Michaud F. A new predic- Cyb. 1990; 20.
tive thermal sensation index of human response. 21 Lee CC, Fuzzy logic in control systems: fuzzy
Energy and Buildings 1999; 29: 167–78. logic controller part 2. IEEE Trans. System Man.
5 ASHRAE Fundamentals, chap. 8, ASHRAE, Cyb. 1990; 20.
Atlanta, GA, 8.1–8.29, 1993. 22 Gouda MM, EL-Ghotmy M, EL-Rabaie N,
6 International Standard ISO 7730. Moderate ther- Sharaf MM. Fuzzy logic control for a turbogen-
mal comfort control environments – erator, Proceedings 5th International Conference
Determination of the PMV and PPD indices and on Artificial Intelligence Applications (ICAIA),
specification of the conditions for thermal Cairo, Egypt, 1997; 1: 592–601.
comfort, 1994. 23 Gouda MM, Danaher S, Underwood CP. Fuzzy

Downloaded from bse.sagepub.com at NANYANG TECH UNIV LIBRARY on April 26, 2015
252 Fuzzy logic controller

Figure A1 Model realization for the internal relative humidity

•
logic controller versus conventional PID M Hr = Ql + mo(Ho – Hr) (A1-1)
controller for controlling indoor temperature of a
building space. Computer Aided Control System The outdoor air enthalpy (Ho) is calculated using
Design (CACSD 2000) Conference. Salford the following set of equations:16
University, UK, 2000.
24 Mamdani EH. Application of fuzzy algorithms 0.624 Pwsat (To )
for control of simple dynamic plant. Inst. Elect. gss (To ) = (A1-2)
Eng. Contr. Sci. 1974; 121: 1585–8. 101.325 - 1.004 Pwsat (To )
25 Sugeno M, Takagi T. Fuzzy identification of
systems and its applications to modeling and Where Pwsat is obtained from either equation A1-
control. IEEE Trans. Syst. Man. Cyb. 1985; 3 or A1-4:
SMC-15, 116–32. The saturation pressure Pwsat over ice (–100 ²
26 Pedrycz W. Fuzzy control and fuzzy systems. T < 0):
USA: JohnWiley & Sons Inc, 1989.
5
ln( Pwsat ) = å ai T i -1 + a6 ln(T ) (A1-3)
Appendix A i=0

To calculate the relative humidity of a building in which,

space, the internal air enthalpy (Hr) is first calcu-
lated from a room air total heat balance as a0 = –5674.5359 a1 = 6.3925247
follows: a2 = –0.9677843 ´ 10–2

Downloaded from bse.sagepub.com at NANYANG TECH UNIV LIBRARY on April 26, 2015
 MM Gouda, S Danaher and CP Underwood 253

a3 = 0.622157 ´ 10–6 a4 = 0.20748 ´ 10–8 Hence the percentage saturation (to which the
a5 = 0.948402 ´ 10–12 a6 = 4.1635019. relative humidity may be taken as a close approx-
imation) is obtained:
The saturation pressure over water (0 ² T ² 200):
fi = 100g(Ti) / gss(Ti) (A1-8)
3
i -1
ln( Pwsat ) = å bi T + b4 ln(T ) (A1-4)
i =-1
Appendix B
Where:
b–1 = –5800.2206 b0 = 1.3914993 The thermal comfort indices (PMV and PPD) are
b1 = –0.048640239 b2 = 0.41764768 ´ 10–4 calculated as follows:1
b3 = –0.14452093 ´ 10–7 b4 = 6.5459673. PMV = (0.352 exp(–0.42Met) + 0.032)
From the model-input value of external air relative ´ (Met – 0.35(43 – 0.061Met – pv)
humidity (fext), the mass of water vapour of the – 0.42(Met – 50) – 0.0023Met(44
mixture (g) of the external air is obtained from: – pv) – 0.0014Met(34 – Ti)
– 3.4 ´ 10–8 fcl((Tcl + 273)4 – Tmrt
fext = 100g(To) / gss(To) (A1-5) + 273)4 – fclhc(Tcl – Ti)) (B1-1)
Thus the enthalpy of the outdoor air (Ho) is given Where Tcl is calculated iteratively from the
by: following equation:
H0 = hg(To) + CTo (A1-6) Tcl = 35.7 – 0.032Met – 0.18Icl(3.4
´ fcl((Tcl + 273)4 – Tmrt + 273)4)
The mass of water vapour in the internal air + fclhc(Tcl – Ti)) (B1-2)
(g(Ti))is given by:
PPD = 1 – 0.95 exp(–0.003353PMV4
g(Ti) = (Hr – CTi) / h (A1-7) – 0.2179PMV2) (B1-3)

Downloaded from bse.sagepub.com at NANYANG TECH UNIV LIBRARY on April 26, 2015