Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Developing environmental controls using a data-driven approach for enhancing environmental comfort and energy performance
(USC Thesis Other)
Developing environmental controls using a data-driven approach for enhancing environmental comfort and energy performance
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
Developing Environmental Controls Using a Data-Driven Approach for Enhancing Environmental
Comfort and Energy Performance
by
LI-CHEN CHEN
SCHOOL OF ARCHITECTURE
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF BUILDING SCIENCE
AUGUST 2018
1
COMMITTEE
CHAIR:
Joon-Ho Choi, (confirmed)
Assistant Professor
USC School of Architecture
joonhoch@usc.edu
(213)740-4576
COMMITTEE MEMBER #2:
Marc Schiler
Professor
USC School of Architecture
marcs@usc.edu
(213)740-4591
COMMITTEE MEMBER #3:
Shrikanth (Shri) S. Narayanan
Professor
shri@sipi.usc.edu
(213)740-6432
2
ABSTRACT
Nowadays, there are various choices of building control strategies for improving thermal comfort while saving energy.
Using real time simulation to control building heating ventilation, air conditioning (HVAC) system has been studied
in the building science realm for many years. However, the following two aspects of the building control have not
been studied thoroughly in research. One i combining optimization methods to control the set-points on HVAC to
minimize thermal stress for multi-occupant office and the energy consumption, and the other is generating trade-off
lines between thermal stress and energy use. This research integrates Matlab, Python and EnergyPlus with a Data-
Driven Rule-Based Control (DRBC) algorithm method to predict optimal building control strategies.
This paper proposes an overall thermal comfort model by undergoing experiments. The human thermal comfort
experiments were conducted to collect occupants’ thermal preference by surveying their thermal comfort feedback
under different environmental conditions in an environment chamber. The thermal comfort and energy saving model
is proposed to be the objective set-point selecting. Moreover, sensors are implemented in the chamber to collect
individual indoor thermal condition data, such as indoor air temperature, relative humidity, ventilation rate, mean
radiant temperature. DBRC is introduced to find the optimum HVAC dynamic control set points, ventilation mode,
temperature and wind speed, of the diffuser. In addition, the final decision will be affected by both the energy saving
and the thermal comfort of set points optimization process.
The results represent that the use of Energyplus coupled with DBRC for the application of HVAC set-point control.
Besides, albeit the selected set-point combination of HVAC control strategies is not guaranteed to be the best, it will
be at least satisfied the most occupants in an area. Finally, DBRC set-point is possible to improve the efficiency over
ASHERAE standard while achieving occupants’ thermal comfort.
HYPOTHESIS
It is possible to improve energy efficiency over ASHERAE standard comfort zone by using a Data-Driven Rule-Based
Control (DRBC) Approach.
3
TABLE OF CONTENTS
ABSTRACT ..................................................................................................................................................2
HYPOTHESIS .............................................................................................................................................2
CHAPTER 1: Introduction ....................................................................................................................10
1.1 Thermal Comfort .......................................................................................................................10
1.1.1 Thermal Comfort ..................................................................................................................10
1.1.2 PMV model ...........................................................................................................................12
1.1.3 PPD model ...........................................................................................................................13
1.1.4 Adaptive thermal comfort model ..........................................................................................13
1.1.5 Thermal condition consolidated indices ..............................................................................14
1.2 Building Energy Simulation ......................................................................................................16
1.2.1 What is building simulation? ...............................................................................................16
1.2.2 Weather data for simulation ................................................................................................16
1.2.3 Occupancy schedule for simulation .....................................................................................17
1.2.4 EnergyPlus ...........................................................................................................................17
1.2.5 IES-VE ..................................................................................................................................17
1.2.6 Real-Time Simulation ...........................................................................................................18
1.3 Intelligent Control and Optimization Methods .......................................................................18
1.3.1 Decision Trees (DTs) ...........................................................................................................18
1.3.2 Fuzzy Logic Controllers (FLCs) ..........................................................................................20
4
1.3.3 Artificial Neural Networks (ANNs) ......................................................................................20
1.3.4 Genetic Algorithms (GAs) ....................................................................................................21
1.4 Goals and Objectives .................................................................................................................23
1.5 Chapter Summary .....................................................................................................................23
CHAPTER 2: Previous Work: Background and Literature Review .................................................24
2.1 Thermal Comfort Control ...............................................................................................................24
2.2 Building Energy Simulation Study .................................................................................................25
2.3 Optimization Methods and Artificial Intelligence Control ..........................................................26
2.3.1 Genetic Algorithms .....................................................................................................................26
2.3.2 Artificial Neural Networks ..........................................................................................................27
2.3.3 Other Control and Optimization Methods ..................................................................................27
2.4 Summary ...........................................................................................................................................28
CHAPTER 3: Methodology ...................................................................................................................29
3.1 Experiment (Data Collecting) .........................................................................................................29
3.1.1 Climate Chamber Introduction ...................................................................................................30
3.1.2 Heating and Cooling System .......................................................................................................31
3.1.3 Single Occupancy Condition (SOC) Experiment ........................................................................31
3.2 Experimental Data Pre-process ......................................................................................................35
3.2.1 Thermal Environmental Condition Index ...................................................................................35
3.2.2 Thermal Comfort Profile Data ...................................................................................................35
5
3.3 HVAC Control Algorithm - Data-Driven Rule-Based Control (DBRC) ....................................38
3.4 Building Energy Simulation (BEM) ...............................................................................................40
CHAPTER 4: Results .............................................................................................................................41
4.1 Overview of Dataset .........................................................................................................................41
4.2 Indoor Environmental Data ............................................................................................................42
4.2.1 Air Temperature ..........................................................................................................................42
4.2.2 Relative Humidity ........................................................................................................................44
4.3 Industry Recommended Temperature/Humidity Range .............................................................45
4.4 Thermal Comfort Data ....................................................................................................................45
4.4.1 Overall Subjects’ Data ................................................................................................................45
4.4.2 Individual Experimental Data .....................................................................................................46
4.4.3 Data Statistical Analysis .............................................................................................................49
4.5 Summary ...........................................................................................................................................50
CHAPTER 5: Data Analysis and Discussion .......................................................................................51
5.1 DRBC Development .........................................................................................................................51
5.2 DRBC for Four Multi-Occupancy Scenarios under Three-Occupant Condition ......................55
5.3 Six-Occupant Condition ..................................................................................................................57
5.4 Nine-Occupant Condition ...............................................................................................................59
CHAPTER 6: Conclusion ......................................................................................................................61
6.1. Experimental Findings and DRBC Performance of different MOCs ........................................61
6
6.2. Limitation of the work ....................................................................................................................62
6.3. Future work .....................................................................................................................................62
6.4 Summary ...........................................................................................................................................63
REFERENCE .............................................................................................................................................64
7
TABLE OF FIGURES/TABLES
Figure 1 PMV-PPD Chart ............................................................................................................................13
Figure 2 Adaptive Chart ...............................................................................................................................14
Figure 3 Decision Tree Example .................................................................................................................19
Figure 4 Fuzzy Logic Control Flow Chart ...................................................................................................20
Figure 5 Simple ANN Concept ....................................................................................................................21
Figure 6 ANN Process Flow ........................................................................................................................21
Figure 7 GA Flow Chart ..............................................................................................................................22
Figure 8 Methodology Flow Chart Diagram ...............................................................................................29
Figure 9 Climate Chamber Layout ..............................................................................................................30
Figure 10 Climate Chamber .........................................................................................................................30
Figure 11 Tripod Environmental Condition Sensor .....................................................................................32
Figure 12 Experiment Process .....................................................................................................................33
Figure 13 LabVIEW Interface .....................................................................................................................33
Figure 14 Questionnaire ...............................................................................................................................34
Figure 15 MATLAB Curve Fitting Toolbox ...............................................................................................36
Figure 16 RapidMiner's Software Normalize Operator ...............................................................................36
Figure 17 Example of Thermal Profile ........................................................................................................37
Figure 18 2-Occupancy Example .................................................................................................................38
Figure 19 DBRC Classification Diagram ....................................................................................................39
8
Figure 20 Four Scenarios for Classification (3-Occupancy) .......................................................................40
Figure 21 Temperature increasing procedure ..............................................................................................43
Figure 22 Temperature at four heights .........................................................................................................44
Figure 23 Relationship of T
a
and RH - Experiment for Subject A&B ........................................................44
Figure 24 CBE Thermal Comfort Tool ........................................................................................................45
Figure 25 Overall thermal comfort ..............................................................................................................46
Figure 26 Percentages of Subjects Above the Neutral Line ........................................................................46
Figure 27 Thermal Comfort Profile - Subject A ..........................................................................................47
Figure 28 Thermal Comfort Profile - Subject D ..........................................................................................47
Figure 29 Thermal Comfort Profile - Subject B ..........................................................................................48
Figure 30 Thermal Comfort Profile - Subject F ...........................................................................................48
Figure 31 Thermal Comfort Profile - Subject G ..........................................................................................49
Figure 32 Data analysis of the whole dataset - Gender ...............................................................................49
Figure 33 Data analysis of the whole dataset – BMI ...................................................................................50
Figure 34 DRBC structure ...........................................................................................................................51
Figure 35 Thermal comfort profile function code for Subject A, Subject B and Subject G .......................52
Figure 36 Three-occupant condition (Subject A-B-G) ................................................................................52
Figure 37 Python flowchart diagram ...........................................................................................................53
Figure 38 DRBC model in Python ...............................................................................................................54
Figure 39 Four Scenarios MOC ...................................................................................................................55
9
Figure 40 Three-occupant condition (Subject D-F-H) .................................................................................56
Figure 41 Three-occupant condition (Subject T*-B-O) ...............................................................................56
Figure 42 Six-occupant condition S#1 & S#2 .............................................................................................57
Figure 43 Six-occupant condition S#3 .........................................................................................................58
Figure 44 Nine-occupant condition S#1 & S#2 ...........................................................................................59
Figure 45 Nine-occupant condition S#3 ......................................................................................................60
Table 1 Metabolic rates for resting and office activities ..............................................................................11
Table 2 Clothing level ..................................................................................................................................11
Table 3 Thermal Sensation Scale .................................................................................................................12
Table 4 GA’s Algorithm Structure (G. Zhang et al., 2107) .........................................................................22
Table 5 Results of Han's research ................................................................................................................25
Table 6 Yiqun's Research Results for Building A and B .............................................................................26
Table 8 Environmental data records ............................................................................................................41
Table 9 Survey data records .........................................................................................................................41
Table 10 Dataset information .......................................................................................................................42
Table 11 Temperature increasing regression trend line equation ................................................................43
Table 12 TCL value of selected subjects under the computed set-point condition (6OC) ..........................58
Table 13 TCL value of selected subjects under the computed set-point condition (9OC) ..........................60
10
CHAPTER 1: Introduction
The control of heating, ventilation and air conditioning (HVAC) systems has been facing the trade-off between
providing acceptable thermal comfort conditions and reducing the energy consumption related to HVAC systems. The
aim of this research is establishing a data-driven rule-based HVAC systems control (DRBC) method. DRBC integrates
individual occupants’ thermal comfort preferences into the control logic which can contribute to solve that conflict
between thermal comfort and energy saving.
This chapter introduces and discusses the technical terminologies related to thermal comfort, building energy
simulation (BEM) and intelligent control and optimization methods.
1.1 Thermal Comfort
1.1.1 Thermal Comfort
As defined by the ANSI/ASHRAE Standard 55-2017, thermal comfort is “the condition of mind which expresses
satisfaction with the thermal environment and is assessed by subjective evaluation.” Thermal comfort is the occupants’
satisfaction and sensation with the surrounding thermal conditions and is necessary to be considered when designing
a building.
Energy-efficient buildings are sitting on top of the basic comfort level of occupants in the buildings. If occupants are
not comfortable, then alternative strategies of heating or cooling a space such as space heaters or portable air
conditioners will be taken and could be considerably worse than normal/typical HVAC systems. Therefore, the thermal
comfort should always be considered while designing building energy saving strategies. If no such actions are taken,
the comfort and productivity of the occupants may suffer.
The thermal comfort is particularly subjective. This makes thermal comfort hard to be quantified and measured. There
are factors which directly affect the human thermal comfort level: the air temperature, humidity, radiant temperature,
air velocity, clothing levels, and metabolic rates and everyone has different thermal sensations based on his or her
physical and psychological state. For example, a warm sensation will be pleasing when the body is overcooled, but
unpleasant when the core is already hot. In addition, the skin temperature is not uniform on all parts of the body. The
clothing level also has a huge effect on the level and distribution of body temperature. Thus, the thermal sensation
from any of the skin on any part of the body will depend on activity, location and clothing, as well as the thermal
condition of the environment (2017 ASHRAE Handbook - Fundamentals).
There are six determining factors affecting the thermal comfort: metabolic rate (met), clothing insulation (clo), air
temperature (°C), radiant temperature (°C), air velocity (m/s), relative humidity (%). These factors can be sorted into
two categories, which are environmental factors and personal factors. The environmental factors include temperature,
radiant temperature, relative humidity, and air velocity. The personal factors are activity level (metabolic rate) and
clothing, as shown as Table 1 and Table 2.
11
Table 1 Metabolic rates for resting and office activities
Table 2 Clothing level
Thermal comfort can be regarded as a balance between heat/energy transfer and the occupant’s metabolic rate. The
heat transfer through radiation, conduction and convection occurs between the environment and the human body. To
be more specific, the thermal perception is “warm” or “hot” if the heat entering the occupant is greater than the heat
leaving the occupant. On the other hand, the occupant perceives “cold” if the heat loss of the occupant is greater than
the heat gain of the occupant. However, since thermal comfort is highly a state of mind, the thermal sensation and
satisfaction are still subjective and difficult to quantify.
A most widely known method of describing thermal comfort was developed by Ole Fanger in 1970 and is referred to
as Predicted Mean Vote (PMV), Predicted Percentage of Dissatisfied (PPD) and Adaptive Comfort Model. Based on
the model above, ASHRAE has developed a consensus standard, as known as ASHRAE Standard 55-2017 which is
frequently used in this research (2017 ASHRAE Handbook - Fundamentals). In addition, the international thermal
comfort standard ISO 7730 is also based on the PMV/PPD model and were commonly used in determining occupants’
thermal comfort level in offices and residential buildings.
12
1.1.2 PMV model
The Predicted Mean Vote (PMV) is an index which represented by a 7-point scale to predict the mean thermal
sensation for a group of occupants, originally defined by Ole Fanger (1970) and later adopted as ISO and ASHRAE
thermal comfort standard. The PMV index is calculated by considering four environmental variables (air temperature,
air velocity, humidity, mean radiant temperature) and two personal variables (metabolic rate and clothing insulation).
In addition, the PMV model is developed based on an extensive experiment, which collected the thermal sensation
data from many subjects to different thermal conditions within a climate chamber. In the experiment, the subjects
were asked to rate their own comfort level on the 7-point bipolar scale (from -3 to +3), as shown in Table 3, to best
describe their thermal sensation. With the experiment data, the PMV model was established by finding out the
relationship between the environmental and physiological factors. Therefore, the PMV model is a tenable and useful
tool, which can be utilized in various interior spaces with different thermal condition and human behavior, such as
HVAC system, clothing and activity (P.O. Fanger, 1970).
Table 3 Thermal Sensation Scale
Furthermore, Fanger concludes that thermal comfort is achieved when the body and the environment is in a heat
balance. Through the heat balance principle, the equation was developed to describe the disequilibrium between the
heat flow in and out from a human body and the heat flow required to maintain comfort for an activity and environment
(D. Enescu, 2017):
PMV = [0.303 · exp(−0.036) · M + 0.028] · L
M: Metabolic rate
L: Thermal load which is defined as the difference between the internal heat production and the heat loss to the actual
environment for a person hypothetically kept at comfort values of skin temperature and evaporative heat loss by
sweating at the actual activity level.
Apparently, zero is the ideal value which represents thermal neutrality and the recommended acceptable PMV thermal
comfort range from ASHRAE 55-2017, EN 15251 and ISO 7730 is between -0.5 and +0.5 for the indoor environment.
Moreover, ASHRAE Standard 55-2017 uses the PMV model to set the requirements for indoor thermal conditions. It
requires that at least 80% of the occupants should be satisfied with the thermal condition of the interior space (2017
ASHRAE Handbook - Fundamentals).
13
1.1.3 PPD model
Fanger (1970) established another equation, Percentage of Dissatisfied to the Predicted (PPD), which is related to the
PMV as follow:
PPD = 100 – 95 · exp[-(0.03353 · PMV
4
+ 0.2179 · PMV
2
)]
The empirical relationship between PMV and PPD is shown in Figure 1 (Olesen, 1982). PPD is a quantitative measure
of the thermal comfort of a group of people at a thermal environment. Similar to PMV, it was also based on the human
subject experiment in a climate chamber. PPD forecasts the percentage of occupants that will be dissatisfied with the
different thermal conditions. The maximum number of people dissatisfied with their thermal conditions is 100%. It
implies that it is nearly impossible that all the people can be pleased all the time. The recommended acceptable PPD
range for thermal comfort from ASHRAE 55 is less than 10% of the occupants are dissatisfied with the indoor thermal
indoor environment.
Figure 1 PMV-PPD Chart
1.1.4 Adaptive thermal comfort model
The adaptive comfort model considers more human behavior factors. Van Hoof (2008) suggested that human thermal
perception is capable to adapt to local climate and their thermal comfort neutral temperature is much dependent to
outdoor long-term averaged ambient temperature when indoor space is mainly operating in natural ventilation mode.
Adaptive thermal comfort assumes that a human reacts differently in order to maintain their comfort level if the change
of the environment produces discomfort. That is, people take actions and make physical adjustments to achieve
thermal comfort or a level of satisfaction by changing their body’s heat balance. For instance, changing clothing,
reducing activity levels, opening/closing a window/door, etc. Based on this assumption, humans can adapt to different
temperatures in a certain range: the comfort temperatures vary depending on outdoor climate, such as higher indoor
temperatures are allowed at higher outdoor temperatures.
The adaptive model is developed for evaluating the indoor thermal comfort under natural ventilation conditions by A.
Auliciems (1981); R. de Dear and G. Brager (1997) and F. Nicol, M. Humphreys (2002). The model is represented as
equation below. The adaptive model was also adopted by ASHRAE as standard for evaluating thermal comfort in free
running buildings.
T
"#
= T
"%&'
+T
"%&)
+⋯+T
"%&+,
×30
&'
T
0 123415
= 0.31∙T
"#
+17.8
14
Where T
om
is monthly average temperature in the previous 30 days; T
od-i
is daily average temperature i days before the
calculated day; T
n
(ASHRAE)
denotes to thermal adaptive model’s T
n
proposed by ASHRAE.
Therefore, to consider adaptive comfort, the space should be naturally ventilated which have operable windows, no
mechanical cooling system, and the occupants’ activity level is constrained: restoring a metabolic rate between 1.0
and 1.3 met. Different from PMV-PPD model, occupants are allowed to add or remove their clothing to adapt to the
thermal conditions. According to the Figure 2, the adaptive chart, proposed by ASHRAE, relates indoor comfort
temperature to prevailing outdoor temperature and defines zones of 80% and 90% satisfaction. Thus, the adaptive
model has a wider range of thermal comfort conditions that occupants can be considered as comfortable. (ASHRAE
Standard 55-2017).
Figure 2 Adaptive Chart
1.1.5 Thermal condition consolidated indices
It is difficult to show a thermal condition in an area based solely on the indoor air temperature or humidity. Thus, it is
necessary to develop an index which combines different parameters to evaluate the thermal condition. For a
quantitative assessment of exposing people to a certain environment, thermal condition indices integrate various
thermal environmental parameters into a single number. Thermal stress indices can involve direct measurements of
environmental parameters such as dry-bulb temperature, wet-bulb temperature, relative humidity, airflow velocity,
etc. Over 160 indices have been proposed for evaluating different environments since 1905 (de Freitas and Grigorieva,
2014). Additionally, the aggregated indices are widely used in both industrial and researching field. Some of them are
introduced in this section:
Operative Temperature (OT)
The operative temperature (OT) was proposed by Winslow, Herrington and Gagge in 1937 (Winslow, Herrington and
Gagge, 1937). It is defined as “the uniform temperature of an imaginary black enclosure in which an occupant would
exchange the same amount of heat by radiation plus convection as in the actual non-uniform environment” (ASHRAE
Standard 55-2017).
OT is obtained from air temperature, mean radiant temperature (MRT) and air speed (ISO 7726). It can also be the
average of MRT and dry bulb temperature (DBT), which is the air temperature but blocked from moisture and radiation,
weighted by their respective transfer coefficients. The expression of OT is:
OT = MRT + t
?
× 10v × (1 + 10v)
&'
OT = (MRT × h
D
+ DBT × h
G
) × (h
D
+ h
G
)
&'
Where t
a
is air temperature (°C), v is air velocity (m/s), and h
r
and h
c
are radiation and convection coefficients
respectively.
15
Wet Bulb Globe Temperature (WBGT)
The wet bulb globe temperature (WBGT), which was created by Yaglou and Minard in 1957, is a temperature index
that estimates the combined effect of temperature, humidity, sunlight (sun angle and sun radiation) and wind speed on
people. The WBGT was developed originally for a simple field study of the effective temperature (ET) for the control
the fatalities due to heat in US military academies. After WBGT implementation, there was a decrease in heat-related
illnesses during the military training. The equations of WBGT are the weighted average of wet bulb temperature
(WBT) under natural ventilation, globe temperature (GT), which is measured with a matt black coating hollow copper
sphere to absorb radiant heat and a temperature sensor at its center, and dry bulb temperature (DBT) (Parsons, 2006).
WBGT
outdoor
= 0.7 · WBT + 0.2 · GT + 0.1 · DBT
WBGT
indoor
= 0.7 · WBT + 0.3 · GT
In addition, Blazejczyk (2011) introduced a simplified equation of WBGT in the study, which is written as:
WBGT = 0.567 · DBT + 0.393 · P
a
+ 3.94
Where P
a
(Pa) is the air vapor pressure.
Effective Temperature (ET)
Effective temperature was developed by Houghten and Yagloglou (1923) in which a set of lines of comfort were
drawn on the psychrometric chart. represented by a set of equal comfort lines drawn on the psychrometric chart. It is
defined as “the temperature of a still, saturated atmosphere, which would, in the absence of radiation, produce the
same effect as the atmosphere in question (Vanos et al., 2010).” Thus, it combines the effect of air temperature and
humidity. It became the most prevalent thermal index for the next few decades. ET is defined in equation as:
ET = DBT−0.4 × DBT−10 × (1−
43
',,
)
However, Yaglou (1947) and Glickman et al. (1950) stated that the ET overestimates the effect of humidity under
comfortable conditions, especially at lower temperatures. To correct the deficiency for the index, some indices were
then being proposed to supplant ET, such as new effective temperature (NET).
New Effective Temperature (NET)
The new effective temperature (NET), as same as ET, considers the effect of air temperature, humidity, and air
movement creating an equal thermal sensation. The expression of NET developed by Gagge et al (1971) represents
the effective temperature of the body at various meteorological parameters and it shows good correspondence with
isotherms for skin temperature, skin moisture, heart rate and discomfort votes on psychrometric chart. Besides, NET
is expressed in air temperature, relative humidity and air velocity:
NET =
+J & (+J & K
L
)
, .MN & , .,,'O × 43
+
'
'.JM × '.O × P
Q.RS
& , .)T × K
L
× (' & , .,' × 43 )
Thermal Strain index (TSI)
The thermal strain index was created by Lee DHK. (1958) based on both observation and the heat transfer mechanisms
analysis. TSI was also plotted by a set of equal strain lines on the psychrometric chart. The following is the expression
of TSI:
Thermal Strain Index (TSI) =
'
+
× t
U
+
+
O
× t
?
– 2 × v
, .X
16
Temperature Humidity Index, Thermohygrometric Index (THI) or Discomfort Index (DI)
The temperature humidity index, so called thermohygrometric index (THI) was developed by Thom in 1959 (R.
Bhattacharya, 2014). THI was originally called the discomfort index. It considers both the influence of temperature
and humidity which measures the degree of discomfort of an individual in warm weather. THI can be expressed by:
THI = 0.72 (t
a
+ WBT) + 40.6 or
THI = t
a
− (0.55 − 0.55 · RH) · (t
a
− 5)
1.2 Building Energy Simulation
1.2.1 What is building simulation?
Building simulation plays a more significant role not only in building design, but also in the operation, diagnostics,
commissioning and evaluation of buildings energy and environmental performance in the last two decades. Building
energy simulation (BES) is the software tool to estimate the energy performance of a building. A typical building
energy model will have inputs for climate; envelope; internal gains from lighting, equipment, and occupants; HVAC
equipment and control systems; occupants, equipment, and lighting schedule. Energy models will then output building
energy use in typical end-use categories, such as plug load, lighting, fan, heating, cooling, etc. Building energy
simulation can help the designers compare various design strategies and lead them to optimal and energy saving
designs. It can also help the managers and engineers define the energy saving potentials.
There are many building energy simulation software programs available nowadays, among which, some are simplified
energy analysis models, which are suited to provide a quick analysis the energy use of building’s performance; some
have complex function and better ability to conduct a comprehensive simulation and analysis.
1.2.2 Weather data for simulation
Weather is the primary and initial factor which affects the building energy usage in terms of maintaining the indoor
thermal level. The weather data is undoubtedly an indispensable input for simulating the building energy performance.
Therefore, selecting an appropriate weather data for simulation increases the accuracy and credibility of the building
simulation result. Most of the energy simulation software tools have a wide variety of weather data: weather data
source, locally recorded weather data or preselected 'typical' years, often a considerable range of options. Most of the
weather zones are available to get the typical weather data from the simulation program’s data base. However, some
simulations need to be scrutinized further may have special needs for specific locations. The weather data can be
generated by a self-owned weather station.
In addition, most of the building simulations should avoid using a single year because no single year can represent the
typical long-term weather patterns. More comprehensive methods produce a synthetic year to represent the weather
data, such as temperature, humidity, wind velocity, solar radiation, and other variables. These types of weather data
are more appropriate and can reflect energy use and thermal analysis that are closer to the long-term average. Typical
Meteorological Year 2 (TMY2) and Weather Year for Energy Calculations 2 (WYEC2) are the mostly commonly
used data using this type of method, which are more closely match with the typical climatic conditions of an area.
TMY2 of Los Angeles city were used in this research as the weather data input while doing energy simulation. TMY2
is developed from analyzing multi-year weather data and extract data of the month within the chosen years that most
represent the typical weather type of the season in the area. For example, a TMY2 could be composed of data from
17
January for 1997, February for 1999, March for 2000, etc. Therefore, TMY2 represents the typical weather pattern of
an area. However, TMY2 has a disadvantage of not suitable for simulating extreme weather.
1.2.3 Occupancy schedule for simulation
Most of the structure are designed for human use on the basis of different activities. The number of occupants,
occupants’ activities and habits can considerably influence the equipment use schedule (e.g. lighting schedule,
cooling/heating schedule, natural ventilation schedule, etc) and building energy use. Moreover, occupancy is directly
influencing the internal loads and ventilation requirement, which largely related to building energy consumption. Thus,
an occupancy schedule is one of the primary inputs for building simulation. Occupancy in the building simulation
programs is mostly defined as: the occupancy status of a space, number of occupants in a space, and the space location
of the occupants. A thorough report of occupancy schedule can reflect the real behavior of the occupants in a space.
Some of the simulation software can simulate more detailed occupancy, such as occupants’ features, in buildings to
improve the simulation of energy consumption.
1.2.4 EnergyPlus
EnergyPlus software is the simulation tool which has a relatively powerful calculation engine for the building energy
performance. It is one of the most robust simulation tools available in the world today. Furthermore, it is the energy
modeling engine that the US Department of Energy is betting the future of building modeling on. The DOE-2 engine
that runs eQuest (Quick Energy Simulation Tool) is replaced and its algorithms are improved for modeling bewildering
building designs and mechanical systems. The program is used for building energy analysis and thermal load
simulation. There are several Graphical User Interfaces (GUIs) that developed based on the EnergyPlus engine, such
as OpenStudio and DesignBuilder, etc. In addition, EnergyPlus fully integrated building shell, geometry, HVAC and
renewables simulation. It models building heating, cooling, lighting, ventilating, and other energy flows, as well as
water. The program includes many innovative simulation capabilities, such as time steps of less than an hour, modular
systems and plant integrated with heat balance-based zone simulation, multi-zone air flow, thermal comfort, water
use, natural ventilation, and photovoltaic systems. Perhaps the one of the limitation is that the rectilinear geometries
are suggested. However, a complex geometry can still be built and simulated while all the areas are defined as heat
balance-based zones.
1.2.5 IES-VE
Similar to EnergyPlus, the IES Virtual Environment (VE) is a suite of building energy performance and environmental
analysis modeling applications, which was developed integrated Environmental Solutions Ltd. Its analysis engine,
ApacheSim is capable of 3-D modelling, solar shading simulation, lighting simulation, dynamic thermal simulation,
building load calculations based on ASHRAE’s method, CFD (computational fluid dynamics) modelling, HVAC
controls system simulation, renewable energy systems simulation, water systems simulation, etc. With these
capabilities, it can be used by designers to test different options, identify passive solutions, compare low-carbon &
renewable technologies, and draw conclusions on energy use, CO2 emissions and human thermal comfort. Like
EnergyPlus, VE also contains an integrated central data model, which can directly link to SketchUp, Revit,
Vectorworks and gbXML, IFC & dxf files imports. Moreover, IES-VE has the same heat balance-based zone
calculation logic as EnergyPlus that all zones should be defined and must be enclosed.
18
1.2.6 Real-Time Simulation
In a real-time simulation, the dynamic variables, which are the variables change with time, of computer model change
at the same rate as the physical condition in the real world. For example, if an office has 10 people in the real-world
in real time, the simulation would have 10 people as well. Configuring models to run in real time enable the
controllers be tested by simulation. Design can be changed earlier in the development process, reducing costs and
shortening the design cycle. Unlike the simulation configuring in real-time, most of the simulation is performed in
a discrete time with constant steps also known as fixed step simulation as time moves forward in equal duration of
time. In addition, the time required to solve the internal state equations and functions representing the system must be
less than the fixed step. In simple words, real-time simulation must produce the internal variables and output within
the same length of time parallel as its physical actual world would. Thus, running models in real time enables you to
use hardware-in-the-loop simulation to test controllers. The system can make design changes earlier in the
development process, reducing costs and mitigating the design cycle.
1.3 Intelligent Control and Optimization Methods
In the United States, buildings account for approximately 41% and 40% of the energy consumption and greenhouse
gas emission, respectively (Building Energy Data Book, 2011). Thus, the energy sector comprises a substantial
fraction of the energy consumption globally. Moreover, the energy consumption in the HVAC equipment in all
residential, commercial, and industrial buildings constitutes approximately 40% to 50% of the world's energy
consumption (H. Mirinejad et al, 2012). Therefore, the magnitude and significance of enhanced energy efficiency in
order to reduce the energy usage of HVAC systems through advanced control methods is both pivotal and well-
recognized.
Apart from energy consumption, the thermal comfort of occupants is the most essential element that needs to be
considered while designing and implementing HVAC system. In other words, the indoor thermal comfort of occupants
should be ensured while reducing the energy use on the HVAC systems. Accordingly, many techniques of HVAC
control have been developed based on improving both thermal comfort and energy performance in recent years.
Comparing with conventional controllers, such as On-Off and proportional, integral and derivative (PID) controllers,
intelligent controllers not only can remarkably save the energy use but also improve thermal comfort to occupants in
the building simultaneously (H. Mirinejad et al, 2012).
Few advanced control methods and algorithms (e.g. rule based, knowledge based data-driven, artificial neural network,
genetic algorithms, etc.) are introduced in this section:
1.3.1 Decision Trees (DTs)
Decision trees (DTs) learning is a statistic, data mining and machine learning method. DTs are developed to predict
the value of a target variable (represented in the leaves) based on observation about several input variables (represented
in the branches). A probabilistic decision tree example is shown in the Figure 3. It is a simple example for
representation for classifying. There are mainly two types of DTs: Classification Tree and Regression Tree. For the
classification tree, the target variable can take a discrete set of values and the predicted outcome is the class to which
the data belongs. The DTs where the target variable can take continuous values or real numbers are called regression
trees (Data Mining with Decision Trees: Theory and Applications 2015).
19
Figure 3 Decision Tree Example
20
1.3.2 Fuzzy Logic Controllers (FLCs)
Fuzzy logic (FL) was introduced in 1965 by Lofti. Fuzzy Logic, namely, deals with the problem which its function is
changing continuously and is difficult to be defined as either true or false. Therefore, it is based on the human decision-
making methodology. Rather than true or false logic, it uses a degree of truth to identify input information.
Figure 4 shows the comparison between True/False logic. and FL. FLCs are widely used in many studies in the
application area of HVAC systems control. Since FLCs are nonlinear mapping of real variables input and output data,
FLCs are appropriate for various complex engineering where traditional control methods do not achieve
comparatively-favorable results (R. Alcal'a et al, 2003). Compared to conventional controllers, the main advantage of
FLCs is that no mathematical modeling is involved. Moreover, since the human thermal comfort sensation is
subjective and vague, FL is suited to describe it depending on the thermal comfort level linguistically (Mirinejad et
al., 2008). The pivotal part of a FLC is a Knowledge Base (KB), which is designed based on operational experience
of human expert or based on learning and self-organization methods which do not require a mathematical model of
the system (Guo, P.Y., Z.H. Guang and Z. Bien, 1998). In addition, FLCs provide the implementation of multi-criteria
control strategies (R. Alcal'a et al, 2003). Thus, the FLCs can properly address the challenges inherent in control of
an HVAC system.
Figure 4 Fuzzy Logic Control Flow Chart
1.3.3 Artificial Neural Networks (ANNs)
Artificial neural networks (ANNs) are self-learning models based on an analogy to how the human brain processes
information. Instead of processing through programming, ANNs acquire knowledge by identifying the information
patterns and relationships in data and are trained through experience itself. ANNs are composed of myriads of single
units, artificial neurons or so called processing elements (PE). Every single unit of neurons are connected with
coefficients and are organized in layers to form the neural structure. ANNs are capable of learning from the experience
to predict an optimal solution. The learning process comes from connecting neurons in a network. The simple ANN
flow chart is shown in Figure 5. Each PE has weighted inputs, transfer function (weight) and one output (node value).
The neural network is determined by the transfer functions, which is the learning rules of its neurons. The neurons
(weights) are the adjustable parameters and the weighted sum of the inputs result in the output (activation) of the
neuron. The activation signal is passed through transfer function to produce a single output of the neuron. During the
ANN training, the neurons are developed repeatedly until the prediction error is minimized to a specified value or the
network achieves the defined level of accuracy. A well trained ANNs can predict the outputs by given new inputs
(Agatonovic-Kustrin and Beresford, 2000). A clearer ANN diagram is shown in Figure 6:
21
Figure 5 Simple ANN Concept
Figure 6 ANN Process Flow
The various applications of ANNs can be organized into prediction, modeling, pattern recognition, etc. In addition,
many studies have explored the use of ANNs in the application area of HVAC systems control. ANN with the
knowledge representation of fuzzy logic is the commonly seen method. This method is described by the term of Neuro
Fuzzy System (NFS) and is often used in problems that its objective is minimizing the error between the output of the
fuzzy system and the target value (H. Mirinejad et al, 2012).
1.3.4 Genetic Algorithms (GAs)
Since Genetic algorithms (GAs) were introduced by Holland in 1975, GAs have been used widely to solve
multidimensional optimization problems in a huge variety of application fields. GAs are powerful machine learning
solution finding methods, which are based on the concept of the evolution mechanisms in natural selection and survival
of the fittest. By simulating natural evolution processes, GAs are capable to search the problems, in which the objective
function is discontinuous, non-differentiable, stochastic, directed, or highly parallel, and artificially evolve solutions
to the problem.
To describe GA’s mechanism, the term “gene” represents the parameters/inputs that encode some elements of a
possible solution. A “Chromosome”, which is composed of “genes”, refers to a possible solution to a problem. Each
chromosome has its fitness value based on its performance, competing with other species (candidate solutions). In
natural selection process that mimics biological evolution, the unfit species will be eliminated, while the fit species
survive. Similarly, to generate a group of possible solutions, the solutions which are more fit survive, while the less
fit ones will be eliminated. In general, the survived species (fitter solutions) are selected for reproducing the next
22
generation using the crossover and mutation operator. The algorithm modifies a population of individual species
(solutions) repeatedly.
To be more specific, the GA’s mechanism is composed of three operators: selection, crossover and mutation.
Reproduction is the process of selecting solutions from the current generation that will pass down the better genes to
improve the fitness values of the next generation. Since reproduction procedures select individuals based on their
fitness values, chromosomes with higher fitness values have a higher probability to be selected. Over successive
generations, the chromosomes have higher fitness value artificially replace the less fit chromosomes and the
population continuously evolving until the optimal solution meet the defined stopping criteria. Typical GAs has a flow
chart and a structure, as shown as Table 4 and Figure 7, that can be described and summarized as follows:
Table 4 GA’s Algorithm Structure (G. Zhang et al., 2107)
Figure 7 GA Flow Chart
Step 1. Problem encoding. Step 2. Random generation of initial population P(t). Step 3. Evaluation of the fitness of
each chromosome in the population. Step 4. Selection Q
1
(t) for reproduction. Step 5. Crossover Q
2
(t) and mutation
Q
3
(t). Step 6. Test Q
3
(t) for stopping criteria. Process stops and returns the solution, if satisfied; return to step 3 and
continue P(t + 1) the process, if not satisfied.
The major advantage of GA is: it is capable in the type of situations where the numerical or mathematical models fail.
As it is a remarkable algorithm, which the progress within each iteration can be easily viewed. GA can be used in
various of application areas such as optimization, design, robotics, image processing, machine learning, automatic
programming, etc.
23
1.4 Goals and Objectives
The discussion so far has been about the significance of the HVAC control system and the problems as the thermal
environmental requirement of the multiple occupants’ condition. The current industry standard and PMV model
cannot meet the occupants’ thermal requirement. The goal, therefore, is to develop a control algorithm that can
improve the thermal comfort condition while reducing the energy consumption. The proposed algorithm can help
generate the optimal environmental temperature considering the multiple occupants’ thermal preference in one HVAC
zone on both aspects: energy saving and thermal comfort.
1.5 Chapter Summary
In this chapter, the HVAC system control methods, thermal comfort study and energy simulation are discussed in
detail. The approach of improving thermal comfort and building energy performance using advanced HVAC control
methods is also introduced. Chapter 2 focuses on the background research and literature review. Chapter 3 describes
the experiment procedure, analysis method, the control model development, and Overall Thermal Discomfort index
in detail.
24
CHAPTER 2: Previous Work: Background and Literature Review
2.1 Thermal Comfort Control
Indoor thermal control has been playing a significant role in giving occupants a better thermal environment, and are
the largest energy consumers in buildings (Vakiloroaya et al., 2014). Indoor thermal control techniques became one
of the main challenges of the both indoor thermal comfort and building energy efficiency area. PMV is commonly
used as estimating thermal comfort, which was introduced by Fanger in 1970. Two classical indicators developed by
Fanger, PMV and PPD model, are reviewed in Djongyang’s research (Djongyang et al., 2010) and in Chapter 1, above.
Diana reviews the most commonly used thermal comfort models and indicators and discussing their problems in
control indoor environmental condition referring to energy management in indoor applications. This thesis shows a
comprehensive view on the prediction of the main indicators which were widely used to define the control strategies
appropriately in indoor thermal environments: which are the indoor air temperature and the PMV index. (Diana, 2017)
The design of the set-point temperature has a significant influence on the building system’s energy consumption as
well as the occupants’ thermal comfort. In Tyler’s simulation research, among seven climate zones, a 1°C set-point
temperature expansion can contribute to 10-15% energy saving in an HVAC system (Hoyt, T., Arens, E. and Zhang,
H., 2015). However, there are a set of assumptions made in the PMV model, which is used to quantify the average
thermal comfort of building occupants. These assumptions require that occupants stay in a thermal zone, which is
controlled by a typical centrally HVAC system, and that the system operates based on predefined set points. In addition,
the comfort preferences should remain constant among individual indoor space or group of neighboring spaces with
similar thermal loads. The parameters considered in the PMV model are: metabolic rate, clothing insulation, air
temperature, mean radiant temperature, air speed, and relative humidity. Therefore, it is a complex method and has
limitations because many measurement indices should be known.
Hyesim Han analyzed the major variables of PMV and simplified the parameters to implement the model without
incurring additional cost or installing equipment such as sensors. As the result, the simplified PMV control realizes
7.0% more thermal comfort and 5.6% more energy reduction than those achieved by room temperature control,
especially in the intermittent operation period. As shown as Table 5. (Han, H. et al., 2014). Yet, there are many factors
that could potentially contribute to the thermal comfort preference of every individual, which are not included in PMV
model, such as individual differences (Lenzuni, P. et al., 2009), biological sex differences (Hodakarami, Jamal and
Nasrollahi, Nazanin., 2012) (Choi, Joon-Ho et al. 2010), regional differences (Yu, Jinghua et al., 2009) and medical
environments (Khodakarami, Jamal and Nasrollahi, Nazanin., 2012) These non-linear feature variables cannot use the
PMV model to evaluate, and make the prediction of the human thermal comfort condition different with the actual
condition.
25
Table 5 Results of Han's research
Other thermal comfort evaluation controlling methods were developed in many previous works. N. Kampelis did a
case study on a University building and established the Daily Discomfort Score (DDS) to assess Demand Response
(DR) control. DR refers to the technical, operational and market framework allowing consumers change refers to the
technical, operational and market framework allowing consumers change their power demand in exchange for
financial or other type of rewards. DDS uses operative temperature (OT), which considers mean radiant temperature
(MRT) and air temperature conditions, as the pivotal input parameter in two each thermal zone. The methodology
frame work in the research is: 1. Developing a thermal dynamic simulation model of the chosen building. 2. Energy
cost model implementation considering various energy tariff zones. 3. HVAC DR control analysis based on the criteria
of cost of energy consumption annually and DDS thermal comfort evaluation. The result shows that preconditioning
has better thermal comfort compared to the baseline scenario. In preconditioning, DDS is high during a certain number
of days throughout the year. On the contrary, although the standard set points are used, the baseline scenario thermal
discomfort conditions are frequent in both heating and cooling periods.
The previous works above have developed some useful control systems based on evaluating thermal comfort. However,
a specific control system aimed at considering the customer preferences should be proposed. These models will enable
the progress of the studies referring to energy system analysis and optimization, energy management in smart energy
buildings, as well as demand side management including customer oriented direct control of the thermal units in indoor
environments.
2.2 Building Energy Simulation Study
Building energy simulation (BEM) plays a significant role in building design as well as in the operation,
commissioning and evaluation of buildings since the late 50’s. That is, a large number of applications on BEM have
been studied based on different purposes. Two previous works related to BEM are reviewed in this section.
Yiqun Pan proposed a research of BEM. In the research, the method of calibrated BEM was introduced based on
related previous works and guidelines. The method also analyzed the energy consumption of two high-rise commercial
buildings in Shanghai, China. In addition, the energy data of the buildings were collected as input parameters to build
up models with DOE-2. The output of simulation was analyzed and compared to the measured energy consumption
data to refine and calibrate the building energy models. Furthermore, the energy conservation measures (ECMs) are
26
evaluated based on the calibrated models, including using the strategies of variable speed chilled water pumps to
replace constant variable speed water pumps, using free cooling during winter and mild seasons to replace inefficient
cooling towers with new cooling tower with high efficiency, and decreasing the lighting power densities. To be more
specific, 1. Changing the secondary chilled water pumps and hot water pumps from constant speed into variable speed.
2. Using free cooling in winter and mild seasons. 3. Decreasing the lighting power density from 12W/m
2
into
9.31W/m
2
by increasing the efficiency of lighting system without sacrificing the illumination level in office (500lux).
Energy saving performance is simulated and calculated to find out which ECM is the best option for each building.
Table 6 shows the evaluation results two chosen buildings (Yiqun Pan et al., 2006).
Table 6 Yiqun's Research Results for Building A and B
2.3 Optimization Methods and Artificial Intelligence Control
The application of traditional control methods in buildings has faced an obstacle of reducing the energy use while
satisfying the thermal comfort of occupants. A wide variety advanced controlling methods have been developed. With
the rapid development of the artificial intelligence technologies, researchers made an effort to develop an intelligent
system for the HVAC system, considering both energy conservation and users’ thermal comfort conditions (Huang
and Nelson, 1994) for large buildings such as hotels, office buildings, airport and other commercial buildings that
need the robust HVAC control system to control the building's microclimate environmental conditions while they
aimed to minimize the energy consumptions (Shepherd and Batty, 2003). In this section, applications of HVAC
intelligent controlling methods are introduced sitting on the top of previous studies.
2.3.1 Genetic Algorithms
In the field of HVAC control system, Genetic Algorithms (GA) were proposed to be used in the optimization of the
fuzzy control logic (FLCs). The combination of the genetic algorithms and artificial neural network (ANN) has been
researched more widely in recent years. Moreover, the GA also demonstrates a better performance to optimize the
control system in finding the optimal point considering the energy cost, system operation efficiency, thermal comfort,
demand response and some other related factors.
Benitez used GAs to develop smart tuned FLCs dedicated to the control of HVAC systems concerning energy
performance and indoor comfort requirements. This complex problem has some specific restrictions because of the
large time requirements existing due to the need of considering multiple criteria. Moreover, long computation time
models are required to assess the accuracy of individual. To solve these restrictions, a genetic tuning strategy
considering an efficient multi-objectives and criteria approach was proposed. In the research, accurate models of the
controlled buildings (two real test sites) have been provided by experts. Lastly, simulations and real experiments were
compared determining the effectiveness of the proposed strategy (Jose M. Benitez et al., 2003).
27
Alajmi used GA to find the most appropriate GA set that obtains the optimum solutions in a reasonable computational
time (less numbers of simulations). The results show that population size is the most significant control parameter and
that the crossover probability and mutation rate have insignificant effects on the GA performance (Alajmi, A. and
Wright, J., 2014). Facundo also used GA to improve energy efficiency and thermal comfort in dwellings. A
sophisticated version of the multi-objective GA was applied to determine the optimal building design, which allows
working with categorical and discrete variables, and the objectives were evaluated using the building energy
simulation software EnergyPlus. As a result of the optimization, reductions were achieved of up to 95% fewer heating
degree-hours and 99% fewer cooling degree-hours in the living room, and up to 99% less heating energy consumption
and up to 82% less cooling energy in the bedrooms (Bre, F. and Fachinotti, V., 2017).
2.3.2 Artificial Neural Networks
The artificial neural network (ANN) is one of the most popular machine learning technologies that has been widely
applied in the both research and industry. It is a computational model that mimics the biological processes of decision
making by the human brain and nervous system (McCulloch and Pitts, 1943). The ANN model can address with the
non-linear variables systems or unclear dynamical variables in the systems. Operators don’t need to understand the
principle of these variables and logic, the model will track the patterns of the data and make appropriate predictions,
based on it (Moon and Kim, 2010). The ANN model has two major processes in the application: (1) self- training by
a back-propagation process; (2) calculating output by a feed-forward process with a set of input neurons, hidden
neurons, transfer functions, and weightiness of each connection between neurons. The predictive and adaptive function
of the control system can be achieved based on these two processes. Moon’s research has been conducted on applying
the AIs to control the system, based on the data acquired from perceiving the surrounding environments. The
successful application of AI to intelligent machine learning helps optimize the systems and control algorithm
concerning working performance and satisfaction of the design purpose. In Moon’s study, ANN control algorithm
was used to optimize the thermal comfort and the energy conservation of buildings in the cooling season. Two ANNs
based models were adopted in the control algorithm: The first model was used to make the prediction of the energy
consumption during the unoccupied time of different setback temperature and the second was used to predict the time
required for advanced cooling operation. The ANN model based control algorithm has a better performance on 18.03%
energy saving than the conventional HVAC system. However, this research still has some limitations: the proposed
control algorithm does not provide the solution to the multi-occupancy condition thermal comfort (Moon and Kim,
2010).
The synergy of the ANN technology and evolutionary algorithms optimize the control of the system to overcome the
non-linear features of PMV calculations, time delay, and system uncertainty (Dounis and Manolakis, 2001), (Calvino
et al, 2004), (Singh et al, 2006), (Kolokotsa, 2001). Including fuzzy adaptive control, minimum-power comfort control
(Federspiel and Asada, 1994), and optimal comfort control (Arthur and Gerald, 1998), some advanced control
algorithms have been incorporated in smart control. Some hydronic heating systems have adopted a back-propagation
algorithm based direct neural network controller to optimize the system control, and its control system performance
is better than the performance of the conventional control system (Kanarachos and Geramanis, 1998).
2.3.3 Other Control and Optimization Methods
Nguyen and Le considered the phases of pre-processing (including the choice of the relevant variables, the settings of
objective functions and constraints, the choice of the optimization algorithm and its coupling with the building
simulation program), optimization (through suitable algorithms), and post-processing (interpretation of the results,
and sensitivity analysis). The points reviewed start from the definition of the problems (static or dynamic) and of the
number of objectives (for single- or multi-objective problems), and consider the presence of constraints and the
possible formulation of optimization under uncertainty (Nguyen DT, Le LB., 2014). The review of Evins is focused
on the optimization methods, indicating that thermal comfort is the second objective mainly considered (together with
28
energy) in multi-objective approaches (Evins R., 2013). Shaikh et al. also address computational optimization methods
and simulation tools (Shaikh PH et al., 2013).
Xiao Chen conducted an experimental evaluation of an occupant-feedback driven model predictive control designed
based on a previously developed dynamic thermal sensation model (MPC-DTS), as well as experimental comparison
to an MPC designed based on Fanger’s PMV model (MPC-PMV). Experimental results demonstrated that by directly
utilizing occupants’ AMV as feedback in the real-time control, the MPC-DTS allows to maintain thermal comfort
with much less energy consumption compared to using the MPC-PMV (Chen, X., Wang, Q. and Srebric, J., 2015).
2.4 Summary
Chapter 2 introduced and summarized the different algorithms that have been adopted into the HVAC system controls.
The comparisons of these algorithms and an analysis of limitations of previous research could be seen in this chapter.
After comparing the genetic algorithm, decision tree, fuzzy logic control, and artificial neural network, the pros and
cons of these algorithms were discussed. GA is not data-driven but this research is data-driven; it is hard to control
the model by fuzzy logic; different as ANN, this research is not focusing on finding the factors that influencing the
thermal comfort. Thus, the data-driven rule-based control was developed (DRBC) based on the data feature of the
experiment and the control objective. The occupant-based control algorithm should have the capability to update
dynamically according to the user’s physical data and environmental data.
29
CHAPTER 3: Methodology
For developing an HVAC system control method, single occupancy condition (SOC) experiments were conducted to
collect the indoor thermal environmental data and develop the overall thermal stress index. Subjects’ thermal comfort
preferences were collected by measuring their thermal comfort feedback under different thermal conditions in a
climate chamber. With the human experiment data, the thermal comfort profiles of occupants are mapped out. Based
on the data collected from the experiment, the set-points of the HVAC system for 3, 6 and 9 occupants’ office were
optimized by Data-Driven Rule-Based Control (DBRC) algorithms. Thereafter, the chamber model was built to
simulate the energy use of the HVAC system in EnergyPlus software. The predicted set-points are the input of the
simulation. The simulation results are compared and discussed in Chapter 5. The overall methodology is experiment
– DBRC– simulation. The methodology flow-chart is shown in Figure 8.
The method of experiment is introduced in the next section. In the single occupancy condition, the data from the SOC
experiment results were used to develop the thermal preference model of the human subjects. The 18 subjects’
experiment results were used to process the responding occupants’ DBRC model. A comparison between the industry
standard and the optimization was made on both, thermal comfort and energy saving aspects. Each step in the
methodology flow chart are explained in detail in this chapter.
Figure 8 Methodology Flow Chart Diagram
3.1 Experiment (Data Collecting)
In this research, data collecting sets a foundation for the algorithm developing process. Single occupancy condition
experiments were conducted in the climate chamber, B11, in the basement of Watt Hall. In total, 18 subjects
participated in the single occupancy condition (SOC) experiment. Subjects’ thermal comfort preference was collected
by measuring their thermal comfort feedback under different thermal conditions in the climate chamber. And input is
Temperature, Ventilation rate and Relative humidity. With the result of the experiment, the thermal stress function is
defined to be optimizing objective function. In the climate chamber, the environmental factors that might influence
human thermal comfort were strictly controlled using professional experiment equipment. The MRT is controlled by
the air temperature, which is an input parameter. Based on ASHRAE-55-2017 suggestion that, the air speed may not
be higher than 0.8 m/s and the elevated air speed must below the direct control of the affected occupants and should
be adjustable in steps no greater than 0.15 m/s. Air-speed was controlled to in the range between 0 m/s to 0.8 m/ and
was controlled at 0.1 m/s in general. In terms of the indoor air quality, CO
2
density was within the range of 700 ppm
to 900 ppm. Since the thermal comfort is significantly affect by personal factors, occupant factors were controlled by
the requirement of their clothing condition and level of activity. All subjects were suggested to maintain 1 clo clothing
insulation condition, and 1.0 met metabolic rate with common office work activities. The main environmental factors
are air temperature (from 20°C to 32°C) and relative humidity (from 20% to 60%).
30
3.1.1 Climate Chamber Introduction
The human subjects’ experiments were conducted in the Environmental Chamber, located in the basement of Watt
Hall at the University of Southern California. The chamber is cuboid, with a length of 4.5 meters, width of 2.9 meters,
and height of 2.4 meters. The climate chamber is as shown as Figure 9, Figure 10. The Environmental Chamber is a
concrete-enclosed room. The wall was covered with high thermal insulation foam board. Therefore, the chamber is
hard to be affected by the outdoor climate condition. Moreover, air-conditioners, heaters and humidifier were placed
at the corner of the climate chamber to manipulate the indoor environmental condition. Since the warm exhaust air
will possibly heat up the climate chamber, the exhaust air of the AC unit was released out from the climate chamber
through the round ducts. The office table was placed in the middle of the room for subjects doing seated computer-
based work and so as the indoor thermal environment sensors. The researcher also worked at the table with the
participants to operate the chamber’s air-conditioning system as well as to monitor the experiment.
Figure 9 Climate Chamber Layout
Figure 10 Climate Chamber
31
3.1.2 Heating and Cooling System
The temperature of the Environmental Chamber is regulated by 2 air-conditioning having capacity of 6000 BTU and
15000 BTU respectively and four portable heaters having a capacity of 1200 Watts each. Air-conditioning units and
portable heaters were installed at the 4 corners of the room. The AC units and heaters related to four flexible ducts,
and supplied conditioned air to the 4 diffuser boxes which were located on the two sides of the subjects’ working desk.
In addition, two flexible ducts supplied cooling air and heating air into the diffuser box to ensure the air movement
within the climate chamber stayed uniform.
The AC1 with 6000 BTU was hacked and connected to a wire board, which can be controlled by the LabVIEW
program along with the fan speed and compressor power. The AC2 with 15000 BTU was controlled manually for
cooling the room to the initial experiment temperature (20°C). Portable heaters were plugged into a socket. The socket
was connected with a data acquisition which can be controlled by the LabVIEW program.
3.1.3 Single Occupancy Condition (SOC) Experiment
The thermal control system program in the Environmental Chamber was established in experiment controlling
software, LabVIEW, using an air-conditioning units (for cooling), and 4 heater units. Environmental condition data
were collected by SensorDAQ, which was connected with LabVIEW. On the other hand, the subjects’ personal
information has a level of relationship on the thermal comfort (Choi, Joon-ho et al., 2010). Therefore, personal
information of the experiment subjects, body mass index (BMI), age, gender, ethnicity, and clothing insulation index
were recorded at the beginning of the experiment. Based on the subjects’ experiment data, the thermal comfort profile
for each subject can be plotted by regression line, which can indicate the occupants’ thermal preference. However, the
thermal preference can plausibly change in times. That is, the individual has different thermal perceptions under
different circumstances, such as mood, health condition, time of the day, etc. This experiment aims to collect the
baseline thermal comfort profile of the participants. Therefore, the research was also simplified based on the
assumption which all the occupants were evaluated at their baseline condition. Nevertheless, the fact that thermal
preference of individual changes is discussed in Chapter 6.
The experiment was conducted while all participants are exposed to the same control variable. Furthermore, the
participants were assigned to the experimental setting in random order. Both the experimental setting and the
environmental conditions, including air temperature and relative humidity, are the independent variables of this study.
Dependent variables are categorized into personal and environmental measures. Subjective measures include the
participants’ thermal comfort condition under different environmental conditions, which was collected by the
questionnaire/survey. Environmental measures consist of the air temperature, relative humidity, and indoor air velocity.
Environmental Data Acquisition
All the environmental sensors in the chamber, including four air temperature sensors at different locations, one mean
radiant temperature sensor, one CO
2
sensor, one total volatile organic compounds (TVOC) sensor, one relative
humidity sensor and indoor air velocity meter were installed on a tripod (shown in Figure 11), which was placed in
the center of the room. Air temperature sensors were placed at different heights: 0m, 0.6m, 1.1m and 1.7m, from the
floor. Since the chamber is simulating the real occupants working in an office. The MRT sensor was placed at a height
of 0.6m from the floor, considering the sitting activity level. Moreover, the indoor air velocity meter was installed at
0.1m above the floor. The installation was following by the suggestion of ANSI/ASHRAE-55-2017. The tripod device
was used to measure the environment temperature and relative humidity. LabVIEW 2016 was used to operate the air-
conditioning system and collect the subject’s real-time data. The interval of the data collection is 10 seconds, and the
value of the senor, including heart rate and environmental conditions, were displayed on the LabVIEW’s front panel
interface.
32
Figure 11 Tripod Environmental Condition Sensor
Subjects’ Conditions Control in the Climate Chamber
The air temperature of the climate chamber was cooled down to 20°C initially before the experiment began. To
guarantee the stability of experimental conditions, subjects are asked to fill the personal information form and go
through a tutorial, which explains the experiment process in detail, 20 mins before the experiment. Food and heavy
activities are not allowed during the experiment. Subjects were asked to do some productivity tests during the
experiment, and they also took the thermal discomfort condition survey every 5 mins. The experiment lasted for one
hour, total 12 times thermal comfort survey were collected.
SOC Experiment Design
There were total 18 subjects participated in the human subject experiment. The University of Southern California
Institutional Review Board (IRB) approved the experimental method and process. Prior to the experiment, the
laboratory staff gave a tutorial to each of the subjects. The temperature of the climate chamber was initially conditioned
to 20°C with a uniform air supply. At the same time, the subjects’ basic information was recorded on the recording
paper, which included the cloth condition, body mass index, age, gender and ethnicity of the experiment subjects.
After setting up the system, the experiment began. The heater was turned on and heated air was uniformly supplied
by four diffusers. In other words, the space was heating up gradually, approximately increasing the air temperature in
a rate of 1°C per 5 minutes. Every 5 mins during the heating process, the staff conducted a survey on the subjects
concerning their thermal comfort condition. The entire experiment process is shown in Figure 12.
There are 5 questions in the survey. The first question is about “if the subject willing to override the AC temperature
set-point in this moment? if so, warmer or cooler?”. It is to determine if the temperature acceptable and the air-
conditioning habit of the occupants. The second and the third question are thermal sensation (very hot, warm, slightly
warm, neutral, slightly cool, cool, very cold) and thermal satisfaction, which is reported in a bipolar 7-point scale from
very dissatisfied to very satisfied, respectively. The reason of using both is a person could possibly feel warm but
satisfied. Therefore, both questions are necessary to clarify the comfort zone range. The last two questions are related
to the indoor air quality satisfaction level and the overall satisfaction level of the work place. In addition, the
questionnaire was designed based on suggestion of University of California, Berkeley, Center of the Built
Environment (CBE).
33
During the experiment, the thermal indoor environmental data were recorded by LabVIEW. Moreover, the subjects’
physical data were collected and displayed on the LabVIEW (shown in Figure 13). After 60 minutes, the data as
exported as an Excel file from the LabVIEW.
Figure 12 Experiment Process
Figure 13 LabVIEW Interface
34
Questionnaire of Thermal Sensation and Thermal Satisfaction
ASHRAE-55 proposed the typical 7-point thermal comfort questionnaire as thermal satisfaction (very satisfied, mostly
satisfied, satisfied, neutral, unsatisfied, mostly unsatisfied, very unsatisfied). Not only using a 7-point scale survey, a
thermal sensation survey was used, to collect the subjects’ thermal comfort and thermal sensation condition in the
experiment. The survey included 7 different thermal sensation conditions based on PMV model. The subjects were
asked to rate their own comfort level on the 7-point bipolar scale (from -3 to +3), which are very cold, cool, slightly
cool, neutral, slightly warm, warm, hot. In fact, all the subjects of the experiment were asked to report their thermal
status every 5 minutes, which is answering the 5 questions in the thermal comfort questionnaire. The questionnaire is
shown in Figure 14.
Figure 14 Questionnaire
35
In addition, the data collected from Q3 is the most important data, which indicates the thermal comfort level (TCL)
of an individual. Q2 shows the thermal sensation of the subject. Using both Q2 and Q3 can find out thermal preferences
of the subjects. For example, a person can be satisfied with a thermal sensation. Besides, Q1 indicates the willingness
of the subjects to change the set-point temperature. It is highly related to thermal satisfaction and can be used as an
indicator of the “actual” acceptable set-point range. However, the data collected from Q1 also shows the air
conditioning habit of the subject. For example, a person might be slightly dissatisfied with the temperature set-point
but not willing to change the set-point for some reasons. The last two questions were asked to verify whether the air
quality influence the thermal satisfaction.
3.2 Experimental Data Pre-process
3.2.1 Thermal Environmental Condition Index
It is inaccurate to define a thermal condition in an area based solely on the indoor air temperature or humidity. Thus,
it is necessary to develop an index which combines different parameters to evaluate the thermal condition of an
environment. Several thermal indices are introduced in 1.1.5. Temperature Humidity Index (THI) considers the
influence of both air temperature (t
a
) and relative humidity (RH) while OT considers mean radiant temperature (MRT)
and air velocity (v). To combine two equations, based on THI, t
a
is replaced by OT to add the factors of MRT and
velocity. Therefore, A combined thermal index, operative temperature-humidity index (OTHI) was introduced in the
study. The combined index OTHI can be established found on the measured thermal data: Air temperature (t
a
), relative
humidity (RH), mean radiant temperature (MRT) and air velocity (v). The mathematical equation can be expressed
as:
OT = MRT + t
?
× 10v × (1 + 10v)
&'
THI = t
a
− (0.55 − 0.55 · RH) · (t
a
− 5)
OTHI = [MRT+ t
?
× 10v ×(1+ 10v)
&'
]−(0.55−0.55×RH)×{[MRT+ t
?
× 10v ×(1+ 10v)
&'
] – 5}
3.2.2 Thermal Comfort Profile Data
Data cleaning
Data cleaning is a process of corrupt data detecting and correcting. The incomplete data or missing feature values of
attributes would influence the results of the algorithms model performance. Therefore, it is necessary to perform data
cleaning in the first step. There are three main categories of data cleaning, including filling in missing values of
attributes and identifying outliers and noisy data. Lakshminarayan has proposed methods to fill in the missing values
of attributes. In the research, missing data were filled in by regression method (Lakshminarayan and Samad, 1999).
Based on complete case data from the same feature before, and after 5 minute records, a curve fitting model was
generated and used to predict the missing values. Two curve fitting methods were used to pre-process the data: Non-
linear regression and B-splines interpolation. In addition, the thermal profiles/functions were plotted and generated
using MATLAB curve fitting toolbox. The interface of the MATLAB curve fitting app is shown as Figure 15.
36
Figure 15 MATLAB Curve Fitting Toolbox
(photo source: https://www.mathworks.com/help/curvefit/fit-comparison-in-curve-fitting-app.html)
Additionally, defining the outliers and noisy data are also crucial in data cleaning process. Binning method was used
in defining the outliers and noisy data. Based on the experiment setting as well as human common physical parameter
range, bins were developed to filter the outliers and noisy data: air temperature from 20°C to 32°C, relative humidity
from 20% to 60% and air velocity from 0m/s to 0.2m/s.
After defining the outliers and noises, a data reduction process was used to remove irrelevant attributes, which can
improve the performance of the control algorithm. Some irrelevant or less relevant attributes may cause the control
algorithms to be affected and result in poor prediction performance. Therefore, data reduction is important in data
preprocessing. In this research, the stepwise regression analysis was used to identify useful variables that have
significant contribution to the thermal comfort. Stepwise Regression is used to analyze the correlation among
predictors, and removes the least significant variables as well as adds the most significant variables during each step.
Data Normalization
Data normalization is a commonly used statistics method to convert data/attributes to the same scales. It is a crucial
and necessary phase of data preprocessing before starting the statistics analysis and developing a data based algorithm.
Because the data-driven rule-based control (DRBC) used the experimental data to established the function, therefore
all the data should be scaled to a specific range. In this research, all scale of level of thermal comfort was normalized
to -3 and +3. The normalization process was undergone using the RapidMiner’s software tool by Normalize operator.
The operator setting panel is shown in Fig.
Figure 16 RapidMiner's Software Normalize Operator
37
Thermal Profile Plotting
With the data of thermal satisfaction, which is the value of a 7-point bipolar scale (-3 to 3). A personal thermal comfort
level graph can be mapped out, which the thermal condition index OTHI is the x-axis and thermal satisfaction is the
y-axis. And the curve fitting lines was generated through MATLAB. The schematic diagram is as shown in the Figure
17. For example, the lower limit point of the neutral thermal comfort level locates at the OTHI is 16.4, which is the
thermal condition has t
a
= 23ºC, v = 0.1m/s, MRT = 24ºC and RH = 30%.
Figure 17 Example of Thermal Profile
Mathematically, the thermal comfort profile of subject #1 is expressed as P
1
, which is plotted based on subject A’s
TCL under different thermal conditions (different OTHI) and curve fit into a regression/interpolation line. This
research assumes that every individual has their unique thermal profile based on their thermal preference. The
domains of TCL and OTHI are presented as:
−3 ≤ TCL ≤ 3,TCL ∈ R
12.6 ≤ OTHI ≤ 26.1,OTHI ∈ R
In addition, for evaluating the thermal discomfort level, thermal stress (TS) is defined as:
TS = −TCL
𝑇𝑆 ≥ 0
or 𝑇S = −TCL 𝑤ℎ𝑖𝑙𝑒 𝑇𝐶𝐿 < 0.
TS is also shown on Figure 17. It is basically the difference between the Pn (P of subject n, n ∈ N) and the neutral
line while TCL is a negative value. TS can be described as the level of thermal discomfort.
In the next section (3.3), the indices introduced in this section were used to develop Data-Driven Rule-Based Control
(DBRC) algorithm. The profile lines of occupants were plotted in one scaled TCL-OTHI chart depending on the
occupancy (3, 6 or 9-occupancy). The profile lines were input into DBRC to find the set-points.
38
3.3 HVAC Control Algorithm - Data-Driven Rule-Based Control (DBRC)
Data-Driven Rule-Based Control (DBRC) is processed in various type of set-points to find out which set-point reduces
the overall thermal stress and energy consumption the most of the building. Moreover, EnergyPlus is coupled with the
DBRC algorithm. The former is utilized to analyze the daily energy consumption in the typical cooling and heating
season for improving the optimization of strategies. In addition, the final decision will be affected by both the energy
saving and the thermal comfort of reformation strategies during the building optimizing process. Hence, this study
also tries to find a trade-off line for a collision between two objectives to provide a basis of selection, by means of the
constraint method and weight method in multi-objective programming for selections of reformation strategies.
For the priority, which is reaching a certain thermal comfort level, DBRC was developed to evaluate the multi-
occupancy condition. In this research, three sizes of occupancy situation are evaluated: 3 occupants, 6 occupants and
9 occupants. In order to indicate all the occupants’ thermal comfort condition in the room, simply using the average
of set-points at the maximum thermal comfort level of the occupants is not applicable. Since, the thermal profile
function is non-linear and the average set-point temperature can possibly be fallen out of one’s thermal comfort zone.
Take Figure 18 as an example, in a 2-occupancy condition, two subjects A and B have different preference offset-
points, OTHI 17 and 22 respectively, and the average is 19.5. However, OTHI 19.5 is out of subject B’s thermal
comfort zone. Instead of choosing the average point, selecting the point that subject A and B have the same level of
thermal comfort within the comfort zone is more logically sound. The given example was a simple 2-occupancy
condition. More sophisticated situation is discussed and evaluated with the DBRC algorithm in this chapter and how
it was developed is introduced in Chapter 5.
Figure 18 2-Occupancy Example
Several advanced control methods and their applications are introduced and discussed in 1.3 and 2.3. In this research,
the human thermal comfort data has been collected in the experiment and the thermal comfort profile functions Sn
was plotted by the data. To evaluate the thermal comfort level for multiple occupancy condition (MOC), different Sn
were randomly chosen and put on the same chart. In this study, DBRC is developed with the classification of various
MOC scenarios using the concept of decision tree (DT). DBRC has 2 if-then decision rules. Decision rules were made
for two classes in the entire process, which is shown in Figure 19.
39
Figure 19 DBRC Classification Diagram
In Class 1, the MOC is classified based on the question “Do all the chosen occupants have the TCL ≥ 0 at the same
OTHI?”. In other words, the MOC is categorized by the thermal comfort zone overlapping condition. Answer “Yes”
means all the chosen subjects have a mutual thermal comfort zone. In contrast, “No” means chosen subjects do not
have a mutual thermal comfort zone. Therefore, the first rule is to distinguish the scenarios of overlapping/discrete
situation.
If the MOC has a mutual comfort zone, which is answer “Yes” for Class 1 (C1-Y), the energy use of HVAC is
considered.
Class 2 is basically the set-point determining rule. In Class 2 rule for C1-Y situation, question “Is it a cooling or
heating condition?” was asked to determine the set-point OTHI. If the MOC is in a cooling condition (e.g. typical
summer), the highest point of OTHI is chosen to save the energy use while all the chosen occupants are above the
neutral level of thermal comfort. In contrast, the lowest point in the mutual comfort zone is selected in a heating
condition. Therefore, the energy saving strategy is triggered based on all the occupants have reached a certain level of
thermal comfort.
However, if the MOC does not has a mutual comfort zone, which is answer “No” for Class 1 (C1-N), the thermal
overall thermal comfort should be considered in the priority. In Class 2 for C1-Y situation, the neutral line is reset to
TCL = -1 (Slightly dissatisfied). Slightly thermal discomfort (TCL > -1) can be acceptable for some people, since the
experiment data point out that some people are not willing to override the set-point under slightly dissatisfied situation.
Thus, the comfort range is extended from TCL ≥ 0 to TCL > -1 in C1-N.
The Class 2 question for C1-N is “Do all the chosen occupants have the TCL ≥ -1 at the same OTHI?”. If the answer
is “Yes” (C1-N/C2-Y), the set-point is determined while the OTHI reaches the maximum overall TCL (Max. STCL).
With this rule, a better overall thermal comfort can be approached. On the contrary, if the answer is “No” (C1-N/C2-
N), the set-point is determined while the OTHI reaches the minimum overall TS (Min. STS). This strategy aims to
minimize the overall discomfort level for the group that individual has relatively discrete thermal condition preferences.
Thus, Class 2 is the set-point decision phase, which is based on Class 1 rules. Figure 20 shows the 4 scenarios (C1-
Y/C2-C, C1-Y/C2-H, C1-N/C2-Y and C1-N/C2-N respectively) in the DRBC classification. Furthermore, the set-
point temperature of 4 scenarios for 3-occupancy condition, 3 scenarios for 6-occupancy condition and 2 scenarios for
9-occupany condition were suggested by DBRC. DRBC was developed in a mathematical with Python.
40
Figure 20 Four Scenarios for Classification (3-Occupancy)
To sum up, DRBC is simple to comprehend and interpret while using rules to categorize, since the thermal comfort
problems are fuzzy and subjective. In addition, DRBC has value even with little hard data. Using DRBC to evaluate
MOC, important insights can be generated based on describing various situations and their preferences for outcomes.
Moreover, DRBC allows the addition of new possible scenarios and the combination with other decision techniques.
3.4 Building Energy Simulation (BEM)
Building Energy Simulation and its application was introduced and discussed in 1.2 and 2.2. BEM is used to predict
and analyze building energy use before the building is constructed. Energy modeling is based on a virtual description
of the geometry and building physics (properties); it relies on a simulation engine (such as DOE2) and often a second
software that provides a user interface. Building energy simulation is widely used in the architecture industry.
In this research, BEM aims to compare the daily energy usage for a typical heating and cooling day which use
suggested set-point as an input with the set-point temperature widely use suggested by the industrial standards such
as IS7730 and ASHRAE 55 2017. The building energy model was built in EnergyPlus software based on the geometry
of experimental climate chamber and calibrate with actual SOC experimental data. After calibration, the office model
was modified to 3-occupancy room, 6-occupancy room and 9-occupancy room.
While modeling, the indoor thermal environmental set-points are the control variables/inputs for the energy simulation.
The energy model was simulated based on the weather data (EPW file) of Los Angeles city and the occupancy/lighting
schedule of typical office (8AM to 6PM). Meanwhile, another model input the set-point which PMV model suggested
was simulated with the same weather condition and schedules. The results of both simulations are demonstrated,
compared and discussed in Chapter 5 and 6.
By analyzing the results, the hypothesis of this research that “One can exceed ASHRAE standard thermal comfort or
energy performance by optimizing a set point given specific occupants’ preferences” was verified (proved/rejected).
41
CHAPTER 4: Results
The data and results of the human thermal experiment are presented in this chapter. Moreover, the experimental data
were analyzed preliminarily to generate statistical information and find out the relationships between the results and
human factors. However, the main purpose of the data collecting process is plotting out the thermal comfort profile of
testing subjects. Otherwise speaking, the collected data are the main input of the control algorithm/model DRBC,
which was introduced in 3.3.
4.1 Overview of Dataset
In the data collecting section, eighteen subjects attended the human thermal comfort experiment. All the experiment
data was transported into a human thermal comfort dataset. Table 7 and Table 8 demonstrates the basic information
of environmental data (total number of experimental records, recording intervals, and length of experiment period,
etc.) and survey data (total number of subjects and survey intervals). The total length of an experiment was 75 minutes
and the first 15-minute data, which is the experiment calibration, was not recorded. Two subjects were tested in an
experiment and nine experiments were conducted in the research.
Before the experiment, the subjects’ individual information was collected including Identity (numbering by letter A-
R), Gender, Body Mass Index (BMI), and Race. During the experiment, environmental temperature (including air at
four levels and mean radiant temperature), and environmental relative humidity was detected by environmental
temperature sensors and relative humidity sensor. All the real-time data with corresponding time were transported into
the dataset by the LabVIEW platform. The dataset information is shown in Table 9. In terms of the race composition
of the experiment, the majority is Asian (three Indians, two Koreans, two Japanese and four Chinese). F. P. Ellis
concluded in his research in 1953 that the level of thermal comfort may vary for different races. However, the race
difference was not considered as a main factor in this research. On the other hand, the attributes gender and BMI of
the subjects were analyzed in this chapter. (Ellis, 1953)
Total records 3240 times
Recording interval 10 seconds
Data averaging interval 60 seconds
Experiment calibration 15 minutes
Duration of experiment 60 minutes
Number of experiment 9 times
Table 7 Environmental data records
Total survey 216 times
Total number of subjects 18 people
Survey interval 5 minutes
Duration of Experiment 60 minutes
Table 8 Survey data records
42
Attribute Type Statistics
ID Nominal 18 subjects
Gender Nominal Male: 11 Female: 7
Age Numeric Min: 19 Max: 54
Ethnicity Nominal Asian: 12 White: 4 Black: 2
BMI Numeric Min: 17.8 Max: 31.5
Temperature Numeric Min: 20°C Max: 32°C
Relative Humidity Numeric Min: 20% Max: 60%
Thermal Satisfaction Numeric Min: -3 Max: +3
Table 9 Dataset information
4.2 Indoor Environmental Data
The environmental data includes the air temperature at four heights, mean radiant temperature (MRT), relative
humidity (RH), CO
2
and TVOC concentration level, and air velocity. However, only the air temperature at head-level
while seated (1.2 m) and RH are discussed in this section. During the climate chamber heating process, relative
humidity and air velocity were controlled in a certain range to minimize the impact of humidity and air movement on
environmental subjects’ thermal sensation.
4.2.1 Air Temperature
The air temperature (at the height of 1.2m) data measured in the experiment also represent the possible temperature
set-point of the HVAC system (approximately from 20 ºC to 32ºC). Figure 21 shows the actual heating process of the
experiment and the data was recorded every minute. Ideally, the air temperature should be increase in a constant rate
of 1ºC per five minutes (0.2ºC/sec). Slightly fluctuating of the temperature during the heating process still happened
in every experiment due to some environmental and personal factors such as the timing of adjusting the set-point of
AC and dehumidifier. Nevertheless, the fluctuations were within a narrow range and are negligible. The temperature
increasing procedures of four groups of experiment are displayed as examples in Figure 21.
To be specific, to interpret the trend line of the temperature increasing data, the unit of the x-axis of the equation in
Figure 21 was converted into “seconds” (shown in Table 10). Besides, the trend line was generated by linear regression.
While running the experiment, the designed rate of increasing the air temperature was 0.2ºC/sec. Yet the actual
experiment showed that the rate was distributed between 0.17ºC/sec to 0.2ºC/sec, which was acceptable. Take
“Experiment A&B” (Table 10) as example, the equation can be described as the initial temperature of the experiment
was 21.231ºC and increased 0.1752ºC per second. In addition, no obvious outlying temperature data was found in the
experiment data.
Additionally, Figure 22 shows the temperature differences at four levels (experiment for subject A&B). In general,
the temperature at different heights changed coherently. However, the temperature change (slope) at the lower height
tended to be gentle. The plausible reason is the air stacked at the lower level while the air temperature at the higher
level changed significantly due to the air turbulence.
43
Figure 21 Temperature increasing procedure
Experiment Trend line equation R²
A&B y = 0.1752x + 21.231 0.97339
C&D y = 0.1745x + 21.465 0.96505
E&F y = 0.1764x + 20.6 0.98528
G&H y = 0.1828x + 21.563 0.98613
x: time (second), y: temperature (ºC)
Table 10 Temperature increasing regression trend line equation
44
Figure 22 Temperature at four heights
4.2.2 Relative Humidity
Relative humidity is highly related to the environmental temperature and pressure. As air temperature becomes warmer,
the amount of moisture that the air can hold increases. That is, given a fixed amount of moisture, the RH decreases
when increasing the air temperature. On the other hand, less water vapor is required to reach a high RH at cooler air
temperatures. Thus, conceptually, RH is inversely proportional to air temperature. Take the experiment of subject
A&B as example, Figure 23 shows the relationship between air temperature and RH in the experiment. RH dropped
from 41.47% to 29.93% while the sir temperature increased from 20.18 ºC to 32.37ºC. The experiment ensured RH
was fell within the range of 20% to 60%. Besides, the indoor RH was regulated by controlling the humidifier.
Figure 23 Relationship of T
a
and RH - Experiment for Subject A&B
45
4.3 Industry Recommended Temperature/Humidity Range
Several industrial consensus standards for thermal comfort were introduced in 1.1. CBE Thermal Comfort Tool
(Figure 24), developed by University of California at Berkley, was used to calculated the environmental temperature
under a specific experiment setting. The relative humidity was set to 40%, Metabolic rate was set to 1.1, Clothing
level was set to 0.5clo, and air speed was set to 0.1m/s. The changing operative temperature, to find the boundary of
thermal comfort requirement with 10% PPD, complied with ASHRAE Standard 55-2017. The temperature range was
from 20.9°C to 26.8°C. Moreover, the occupied comfort heating and cooling set-point is 20°C and 24°C respectively.
The average of the range was used as the baseline set-point temperature (23.85°C) input and the energy performance
of baseline set-point was compared with the energy performance of the optimal set-point in Chapter 5.
Figure 24 CBE Thermal Comfort Tool
4.4 Thermal Comfort Data
In this research, thermal comfort was evaluated by the thermal comfort level (TCL) at different indoor environmental
conditions which are represented as operative temperature humidity index (OTHI). Some representative individual
thermal profiles were chosen and displayed in this section. Moreover, a preliminary analysis of overall subjects’ data
was made to take a quick look at the general pattern of the overall thermal comfort level at different indoor
environmental conditions.
4.4.1 Overall Subjects’ Data
From the dataset of 3240 records, overall thermal comfort conditions with corresponding environmental conditions
can be generated (Figure 25). With the relative humidity controlled from 20% to 60%, the environmental temperature
ranging from 20°C to 32°C and the air velocity was controlled to be 0.1 m/s, the thermal comfort condition ranges
from -3 (very dissatisfied) to 3 (very satisfied). As reported by Fig., the thermal comfort level points of 18 subjects
scattered diversely on the map. However, a hazy overall thermal comfort pattern can be observed. The points fell
frequently above the neutral line (TCL>0) within the range of 17.5 < OTHI < 19.5 approximately (shown in Figure
25 and Figure 26), which implies that most of the subjects in the experiment had thermal preferences around that range.
In addition, “17.5 < OTHI < 19.5” represents that the range of set-point temperature is from 23.7 °C to 26.6 °C while
RH is 40% and air velocity is 0.1 m/s.
46
Figure 25 Overall thermal comfort
Figure 26 Percentages of Subjects Above the Neutral Line
4.4.2 Individual Experimental Data
Totally, eighteen subjects participated in the thermal comfort experiment, and each subject had different thermal
preferences under the climate chamber HVAC control. Therefore, each subject’s thermal comfort experiment data was
used to developed their individual thermal comfort preference profile. All subjects’ thermal preference data were
collected by using thermal discomfort level. Below are five subjects’ thermal preference displayed as examples. The
evaluation method of thermal comfort level was illustrated in Chapter 3. The point indicated occupant’s thermal
comfort level under certain environment condition which is represented as OTHI and the regression line is their
thermal comfort profile. Third order polynomial (nonlinear) regression was used to curve fit the data points.
Conceptually, the profile curve should be a concave down second order polynomial function. In the other words, the
profile should be similar as a Gaussian distribution map. However, the third order polynomial (nonlinear) regression
has higher R
2
(closer to 1).
47
According to Figure 27 and Figure 28, the thermal profile curve of Subject A and Subject D is skew to right, which
means that they had warmer thermal preference. On the contrary, Subject B and Subject F preferred cooler
environmental conditions (shown in Figure 29 and Figure 30). Lastly, Subject G (shown in Figure 31) had a relatively
even thermal preference. Therefore, the skewness of a thermal comfort profile curve displays the thermal preference
of an individual.
Figure 27 Thermal Comfort Profile - Subject A
Figure 28 Thermal Comfort Profile - Subject D
48
Figure 29 Thermal Comfort Profile - Subject B
Figure 30 Thermal Comfort Profile - Subject F
49
Figure 31 Thermal Comfort Profile - Subject G
4.4.3 Data Statistical Analysis
Some influences due to the individual differences, such as gender, were discovered by conducting a statistical data
analysis. Bedford (1948) found that in temperate climates, females usually prefer rather warmer conditions than males.
In this research, the thermal comfort data was sorted by gender to determine the percentages of subjects who were
satisfied (TCL ³ 0) with the indoor environmental condition. The comparison between male and female is shown in
Figure 32. The experiment data showed that the female subjects seemed to have a slightly higher tolerance or
preference of warmer environmental temperature than the male subjects do. In contrast, it was found that the male
subjects prefer a cooler environmental temperature. The findings concur with the Bedford’s statement in his 1948
research. However, the sample size was too small to draw the conclusion that females have a higher tolerance of
warmer conditions. The findings could be more tenable if the quantity of the subjects was increased and other factors
(e.g. age and BMI) were controlled.
Figure 32 Data analysis of the whole dataset - Gender
50
Additionally, data was analyzed based on BMI to assess the relationship between BMI and thermal comfort level. All
subjects’ data were sorted into three groups by three ranges of BMI (BMI<19.5, 19.5£BMI£24 and BMI>24, which
represent underweight, standard and overweight). Figure 33 shows that the subjects that were underweight could
endure higher environmental temperature conditions (right skewed) and the overweight subjects preferred cooler
environmental conditions. Similar to the data analysis based on gender, though the relationship observed in this
analysis was obvious, the sample size was not large enough to make the conclusion. Further, the body fat percentage
and muscle mass percentage should be calculated and taken into account since BMI cannot represent the body shape
or type of a person.
Figure 33 Data analysis of the whole dataset – BMI
4.5 Summary
The result of the experiment was presented and analyzed based on various of aspects in this chapter. However, the
focus of the research is to develop a HVAC control system to determine the indoor set-point temperature with different
occupants’ thermal preference as inputs. Thus, the data presented in this Chapter are primary inputs for establishing
the HVAC control algorithm. The development of HVAC control algorithm, Data-Driven Ruled Based Control
(DRBC), is introduced in the next chapter (Chapter5).
51
CHAPTER 5: Data Analysis and Discussion
With the data collected in the experiment and the HVAC control method introduced in Chapter 3, all the materials and
elements were gathered to be analyzed. The introduction of DRBC has been made in Chapter 3 and the
mathematical/computational form of DRBC is revealed in this section. Moreover, three/six/nine subjects’ profile
(different MOC conditions) were chosen from 18 subjects’ data pool randomly to test the DRBC model. The suggested
HVAC set-point temperature based on the chosen MOC condition was found by the DRBC model.
5.1 DRBC Development
According to Figure 19 in 3.3, DBRC is developed with the classification of various MOC scenarios using the decision
if-then/else rules. The rules were made of three IF decisions in the entire process. DRBC can be divided into three
layers: input layer, decision layer and output layer. The thermal comfort profile function of the subjects were generated
using LabVIEW and MATLAB (the profile function examples were shown in Figure 27, Figure 28, Figure 29, etc.).
The profile functions were selected from the 18 subjects’ data pool and input into DRBC model. In the decision layer,
the classification detail was illustrated in 3.3 and the if/then/else blocks correspond to the blocks in Figure 19. Thus,
DRBC was developed to identify which scenario of MOC condition the selected occupants is and calculate the set-
point based on that scenario. After the set-point temperature were determined, the temperature was set into the
EnergyPlus model and simulated its daily energy use.
Figure 34 DRBC structure
52
DRBC model was written in a mathematical form and built in Python programing. In the input layer, the subject’s
profile functions were selected manually. Take three-occupant condition (3OC) as example, Subject A, Subject B and
Subject G were selected. The thermal comfort profile function of Subject A, Subject B and Subject G are shown as f,
g and h respectively in Figure 35. Then, three functions were first examined if they are overlapped while y (or TCL)
> 0. That is, if there exist a range of x (OTHI) that can make f > 0, g > 0, and h > 0. For this example, three functions
were overlapped (shown in Figure 36). Besides, the green area in Figure 36 is the mutual comfort zone. In this case,
the scenario of Subject A-B-G could either be S#1 or S#2 based on the HVAC was in a cooling or heating condition
respectively. During the cooling period, the maximum x value in the area was found, x = 18.4 (T
a
=25.07°C, RH=40%,
v=0.1m/s), to be the suggested OTHI. In contrast, the minimum x value in the area was found to be 17.2 (T
a
=23.2°C,
RH=40%, v=0.1m/s). Thus, through the DRBC the set-point temperature of a 3OC can be determined.
The example (Subject A-B-G) was for scenario 1 (S#1) and scenario 2 (S#2). Scenario 3 (S#3) and scenario 4 (S#4)
happen if the selected subjects have discrete comfort zone. In other words, the selected subjects’ profile functions do
not exist any x for all the selected functions’ value > 0. To differentiate S#2 and S#3, the chosen functions were
classified if they have an overlapped area above the y (or TCL) > -1. If they are overlapped, the situation is classified
into S#2 and the maximum overall TCL (max.STCL) is calculated. Otherwise, it is S#3 and minimum overall TS
(min.STS) is calculated. The DRBC process of Subject A-B-G in python is shown in Figure 38. The method of finding
solution was iterating x from OTHI = 15 to OTHI = 21 (with float increments) in primarily two if/then loops (shown
in Figure 37). Furthermore, the model was adjusted to compute the solution for six-occupant condition and nine-
occupant condition by replacing f, g and h functions in Figure 38. In the next section, the results of other scenarios are
demonstrated and discussed.
Figure 35 Thermal comfort profile function code for Subject A, Subject B and Subject G
Figure 36 Three-occupant condition (Subject A-B-G)
53
Figure 37 Python flowchart diagram
54
Figure 38 DRBC model in Python
55
5.2 DRBC for Four Multi-Occupancy Scenarios under Three-Occupant Condition
Four MOC scenarios’ graphical explanation is shown in Figure 39. The result of scenario 1 and scenario 2 was shown
in 5.1.1 by the example of Subject A-B-G. The result of scenario 3 and scenario 4 are presented in this section. First,
for S#3, Subject D, Subject F and Subject H were selected to find out the set-point temperature by the DRBC. The
threshold was lower from TCL= 0 to TCL= -1 in this case (shown in Figure 40). In this case, max.STCL was calculated
within the area enclosed by P
D
, P
F
, P
H
(the profile function of Subject D Subject F and Subject H) and TCL= -1.
Besides, if the value of the profile function > 0, it was not counted. Through the calculation, the suggested OTHI was
found to be 18.7 (T
a
=25.45°C, RH=40%, v=0.1m/s). Under this set-point condition, Subject H (TCL= 1.126) was
satisfied while Subject D (TCL= -0.005) and Subject F (TCL= -0.114) were slightly dissatisfied with the indoor
thermal condition. Therefore, the set-point temperature for the scenario of Subject D-F-H is 25.45°C and it is higher
than the cooling set-point temperature (24°C) that ASHRAE Standard 55-2017 recommended.
Second, for S#4, min.STS are computed in this situation. Besides, thermal stress (TS) is defined as -TCL while TCL
< 0. The chosen subjects are categorized as S#4 if no overlapping region exists even the threshold has been lowered
to TCL=-1. However, the situation as S#4 could not be found in the dataset (all the subjects). In order to demonstrate
the result of scenario 4, a testing thermal comfort profile Subject T* was created. The testing profile function P
T
* was
reformed from Subject G’s profile function P
G
by shifting P
G
2.5 units of OTHI along the x-axis. S#4 is not a normal
situation. It could happen when one of the occupants has personal health issue. For example, P
T
* represented an
occupant that was sick and he/she felt cold easily. In this condition, the occupants had entirely discrete comfort zones.
Subject T* and two chosen subjects (Subject B and Subject O) were input to DRBC and found out the suggested OTHI
and set-point temperature. As shown as Figure 41, The result of Subject T*-B-O OTHI is 19.7 (T
a
=26.94°C, RH=40%,
v=0.1m/s). Subject T* (TCL= 0.024) and Subject O (TCL= 0.236) were satisfied while Subject B (TCL=-2.026) and
was dissatisfied with the indoor thermal condition. Instead of using OTHI = 19.7 as the set-point, another possible set-
point could be selected manually as OTHI = 19.1 (T
a
=26.05°C, RH=40%, v=0.1m/s). If so, Subject T* (TCL= 0.510)
would be satisfied while Subject B (TCL= -1.257) and Subject O (TCL = -1.064) were slightly dissatisfied. However,
the computing concept of DRBC is minimizing the overall thermal stress (min.STS = 2.026, while OTHI = 19.7). That
is, make most of the occupants in the thermal comfort zone. And the occupants who are out of the comfort zone can
achieve thermal comfort by changing their body’s heat balance. For example, changing clothing, switching seat close
to or away from the diffuser, etc.
Figure 39 Four Scenarios MOC
56
Figure 40 Three-occupant condition (Subject D-F-H)
Figure 41 Three-occupant condition (Subject T*-B-O)
57
5.3 Six-Occupant Condition
This research aims to find out the better set-point temperature for different sizes of MOC. Different sizes of MOC
represent different occupancy scheduled offices. Six-occupant condition (6OC) is considered in this section. The first
6OC model selected Subject A, Subject G, Subject H, Subject K, Subject M and Subject O to compute the set-point
by DRBC. The thermal comfort profile function of selected subjects is shown in Figure 42. Then, six functions were
examined and classified into S#1or S#2. That is, six functions had mutual region above the neutral lune. In this case,
the this 6OC could either be S#1 or S#2 based on the HVAC was in a cooling or heating condition. During the cooling
period, the calculation result was 18.7 (T
a
=25.45°C, RH=40%, v=0.1m/s) and the set-point was 18.1 (T
a
=24.55°C,
RH=40%, v=0.1m/s) during the heating period.
Second, Subject B, Subject D, Subject F, Subject N, Subject Q and Subject R were selected to compute the set-point
and the suggested OTHI was 18.4 (Ta=25°C, RH=40%, v=0.1m/s) (Figure 43). Under this set-point condition, only
Subject D was fallen out of the thermal comfort zone (Table 11). Therefore, the set-point temperature for the scenario
of this 6OC model is 25°C and it falls within the range that ASHRAE Standard 55-2017 recommended (20°C and
24°C for occupied comfort heating and cooling respectively).
Figure 42 Six-occupant condition S#1 & S#2
58
Figure 43 Six-occupant condition S#3
Table 11 TCL value of selected subjects under the computed set-point condition (6OC)
59
5.4 Nine-Occupant Condition
Employing a strategy similar to the six-people occupancy condition, nine-people groups were made, based on the
subjects’ individual data. In the nine-occupant condition (9OC), also three scenarios of MOC were tested. For S#1
and S#2, the computed set-point was OTHI 18.6 (T
a
=25.30°C, RH=40%, v=0.1m/s) during the cooling days and 17.7
(T
a
=23.96°C, RH=40%, v=0.1m/s) during the heating days (shown in Figure 44). Compared with ASHRAE Standard
55-2017 recommended set-point (two sides of the red area in Figure 44): 24°C and 20°C for occupied comfort cooling
and heating respectively, the DRBC set-point is 1.3°C higher than ASHRAE set-point in a cooling period (S#1), which
can provide energy saving. On the other hand, in a heating period, ASHRAE could possibly have less energy demand
since the set-point is 3.96°C lower than the DRBC set-point. However, if ASHRAE standard 55-2017 recommended
set-point (20°C) were used in this S#2 (heating period), 5 people in the group would be dissatisfied (over-cooled).
Therefore, the since the ASHRAE recommended set-point for heating was poor to reach the occupants’ thermal
comfort, the energy performance is not meaningful.
For S#3, since the mutual comfort zone was not found in this MOC group, the TCL threshold was lower to -1. The
OTHI of the mutual acceptable zone (TCL>-1) was from 17.7 (T
a
=23.96°C, RH=40%, v=0.1m/s) to 18.9 (T
a
=25.75°C,
RH=40%, v=0.1m/s). The computed solution through DRBC was the OTHI at 18.4 (T
a
=25°C, RH=40%, v=0.1m/s).
The detail information is shown in Figure 45 and Table 12. In this S#3 for 9OC, DRBC set-point can save 1°C of
cooling energy use compared to the ASHRAE recommended set-point and only one occupant was dissatisfied with
the environmental temperature under DRBC set-point. On the contrary, ASHRAE set-point can potentially save 5°C
of heating energy but 4 people were dissatisfied with the indoor environmental temperature.
Figure 44 Nine-occupant condition S#1 & S#2
60
Figure 45 Nine-occupant condition S#3
Table 12 TCL value of selected subjects under the computed set-point condition (9OC)
61
CHAPTER 6: Conclusion
Eighteen subjects participated in the human thermal comfort experiment, and their thermal preferences were recorded
by surveying their thermal comfort feedback under different indoor environmental conditions. The environment
condition was controlled with the environment temperature increasing from 20°C to 32°C while the relative humidity
(RH) varying between 20% to 60% and air velocity (v) was 0.1m/s. The experimental data were analyzed on the basis
of different human factors. Furthermore, a Data-Driven Rule-Based Control (DRBC) was used to classified their
scenario and determine the recommended temperature for multi-occupants based on their thermal preference data and
environmental condition (RH=40% and v=0.1m/s). Both energy saving and thermal comfort aspects were considered
to make the performance comparison between the DRBC model and the ASHRAE standard 55-2017’s PMV model.
6.1. Experimental Findings and DRBC Performance of different MOCs
As the experimental data for individual occupants were analyzed, different occupancy conditions were considered in
this research. Thermal comfort experiment data was used to generate the thermal comfort profile function for the
subjects and test the DRBC model’s performance.
Experimental data analysis
A rough pattern of the relationship between thermal comfort level and two personal factors (gender and BMI) was
found. The experimental data showed that the female subjects preferred a warmer environmental temperature while
the male subjects preferred a cooler environmental temperature. Additionally, the subjects who had higher BMI
(overweight) tended to prefer cooler temperature and the subjects that were underweight could endure higher
environmental temperature conditions. However, further research, such as increase the sample quantity, is needed to
provide a sound conclusion.
Three-occupancy condition
Three three-people groups were made for multiple occupancy condition research. DRBC was developed to compute
the set-point temperature based on overall thermal comfort level of the zone. In the computing process, the input
thermal comfort profiles were classified into four scenarios. For the scenario 1 and 2, since all the occupants had a
mutual comfort zone, two sides of the comfort zone’s boundary were selected and considered as the set-point based
on the weather type (heating or cooling). The example of S#1 and S#2 showed that the set-point temperature was
25.07°C and 23.2°C for cooling and heating respectively.
In addition, for Scenario 3 and Scenario 4, DRBC computed the set-point temperature by using the concept of
minimizing the overall thermal dissatisfaction. The suggested set-point temperature for Scenario 3 and Scenario 4 was
25.45°C and 26.94°C. The results of 3OCs were higher than the cooling set-point temperature which ASHRAE
standard recommended. However, the result was based on the dataset and the subjects’ preference of thermal condition.
Six-occupancy condition and nine-occupancy condition
The larger group size, six-people group and nine-people group were discussed and the set-point were calculated.
Similar as three-occupancy condition, the research integrated an individual’s thermal comfort profiles based on the
assumption that individual experiment data is applicable in multiple occupancy condition to simulate a group condition.
In the first group of 6OC, the set-point temperature for S#1 was 25.45°C (cooling) and S#2 was 24.55°C (heating).
The second group of 6OC was categorized as S#3 and the computed temperature was 25°C.
62
The result was compared with the ASHRAE standard recommended set-point temperature for typical cooling and
heating days. In a typical cooling season, S#1 could save energy by 1.45°C usage and S#3 could save 1°C usage
compared to the ASHRAE standard recommended set-point temperature (24°C). On the other hand, in a heating season,
the set-point that ASHRAE standard recommended is 20°C which could use less energy than DRBC result for S#2
(by 4.55°C) and S#3 (°C). However, the energy saving was not meaningful if the set-point was not within the mutual
comfort zone or most of the occupants was not satisfied with the set-point temperature at 20°C.
In the nine-people occupancy condition, the group was classified as S#3 and the set-point temperature was 25°C. The
outcome of 9OC group was similar to the 6OC’s S#3 group since their mutual comfort zone were enclosed by the
same group of selected subjects.
6.2. Limitation of the work
This research has two main limitations that need to be mentioned. First, sample size. There were only 18 subjects in
total, attending the experiment. In order to increase the data-driven approach in terms of accuracy and reliability, more
subjects must be considered.
Second, More validation experiment for historic data. Validation experiments should be done to confirm the reliability
of the thermal comfort profile for each individual. According to the improvement effect of adding validating data size,
it is necessary to conduct more historic experiments and strengthen and update the input thermal comfort profiles,
which help increase the accuracy of the DRBC model.
6.3. Future work
This research has been done based on human thermal comfort experiments carried out in a climate chamber. However,
there are other factors that may influence occupants’ thermal comfort. Thus, SOC and MOC experiments should be
conducted in a real office setting and measuring with more type of sensors. The attributes should include real time
weather conditions, surrounding air speed, cloth insulation, MRT, etc. Moreover, there are other machine learning
algorithms that could be used in this research, such as the GA, ANN, support vector machine, deep learning, etc.
Therefore, one direction of future work could focus on comparing the different machine learnings’ performance on
human thermal comfort prediction, and find the optimal machine learning algorithm.
In addition, occupants’ real-time information is the most important preconditions of using the data-driven approach
for thermal environment control. DRBC model require occupants’ individual thermal preference information input to
the model before using it to control the environment condition. In this research, the occupants’ individual thermal
preferences were collected by human thermal comfort experiment; however, it is difficult to execute such an
experiment for each occupant in the real world. Therefore, developing a device that can collect and detect occupants’
personal information, behavior and physical/psychological state in real time dynamically is another direction of future
work.
With the rapid development of portable smart devices such as the smart phone, smart watch, smart glasses, etc., there
are some potential solutions to solve the problem above by using smart devices. Take the smart watch (or smart band)
as an example. A smart watch can be paired with a smart phone to facilitate the transmission of data. An app can be
developed to collect the occupant’s thermal comfort feedback while recording his/her physical state, such as heart rate,
and its corresponding time using the current technology. At the same time, the environment condition (including
temperature and humidity) can be read by a digital thermal stat such as NEST. Thereafter, the occupant’s thermal
preference can be developed by combining the information obtained from the smart watch and the smart thermal stat.
As for the occupancy identification, the GPS function of a smart watch can be used to identify the occupants in an
area. Thus, the more accurate real-time thermal profile can be generated. As for combining the smart device, some
potential solutions have been proposed and can be done in future work.
63
6.4 Summary
In this research, DRBC approach can possibly better performance in terms of both, energy saving as well as thermal
comfort in multi-occupancy condition, especially in a cooling period. From the energy saving aspect, even though the
PMV model showed good performance in certain conditions, it sacrificed the occupants’ thermal comfort. DRBC
models can demonstrate good energy saving performance while maintaining an acceptable thermal comfort condition.
Thus, the hypothesis “It is possible to improve energy efficiency over ASHERAE standard comfort zone by using a
Data-Driven Rule-Based Control (DRBC) Approach.” was proved.
64
REFERENCE
P.O. Fanger, Thermal Comfort, Copenhagen, Denmark: Danish Technical Press; 1970
Han, H., Lee, J., Kim, J., Jang, C. and Jeong, H. (2014). Thermal Comfort Control Based on a Simplified Predicted
Mean Vote index. Energy Procedia, 61, pp.970-974.
Vakiloroaya, V., Samali, B., Fakhar, A. and Pishghadam, K. (2014). A review of different strategies for HVAC energy
saving. Energy Conversion and Management, 77, pp.738-754.
Hoyt, T., Arens, E. and Zhang, H. (2015). Extending air temperature setpoints: Simulated energy savings and design
considerations for new and retrofit buildings. Building and Environment, 88, pp.89-96.
Lenzuni, P., Freda, D., and Del Gaudio, M. 2009. "Classification of Thermal Environments for Comfort Assessment".
Annals of Occupational Hygiene. 53 (4): 325–32.
Wyon, D. P., Andersen, I., and Lundqvist, G. R. 2009. "Spontaneous magnitude estimation of thermal discomfort
during changes in the ambient temperature*". Journal of Hygiene. 70 (2): 203–21.
Khodakarami, Jamal and Nasrollahi, Nazanin. 2012. "Thermal comfort in hospitals – A literature review". Renewable
and Sustainable Energy Reviews. 16 (6): 4071. doi:10.1016/j.rser.2012.03.054
Yu, Jinghua, Yang, Changzhi, Tian, Liwei, and Liao, Dan. 2009. "Evaluation on energy and thermal performance for
residential envelopes in hot summer and cold winter zone of China". Applied Energy. 86 (10): 1970.
Djongyang N, Tchinda R, Njomo D. Thermal comfort: a review paper. Renew Sustain Energy Rev 2010;14(9):2626–
40
Nguyen DT, Le LB. Optimal bidding strategy for microgrids considering renewable energy and building thermal
dynamics. IEEE Trans Smart Grid 2014;5(4):1608–20.
Evins R. A review of computational optimisation methods applied to sustainable building design. Renew Sustain
Energy Rev 2013;22:230–45.
Shaikh PH, Nor NBM, Nallagownden P, Elamvazuthi I, Ibrahim T. A review on optimized control systems for
building energy and comfort management of smart sustainable buildings. Renew Sustain Energy Rev 2014;34:409–
29.
Chen, X., Wang, Q. and Srebric, J. (2015). Model predictive control for indoor thermal comfort and energy
optimization using occupant feedback. Energy and Buildings, 102, pp.357-369.
Moon, Jin Woo, and Jong-Jin Kim. "ANN-based thermal control models for residential buildings." Building and
Environment 45, no. 7 (2010): 1612-1625. Morel, N., Bauer, M., and El-Khoury, Krauss, J. 2000. “Neurobat, a
predictive and adaptive heating control system using artificial neural networks.” International Journal of Solar Energy,
21, 161–201.
Alajmi, A. and Wright, J. (2014). Selecting the most efficient genetic algorithm sets in solving unconstrained building
optimization problem. International Journal of Sustainable Built Environment, 3(1), pp.18-26.
Bre, F. and Fachinotti, V. (2017). A computational multi-objective optimization method to improve energy efficiency
and thermal comfort in dwellings. Energy and Buildings, 154, pp.283-294.
ISO 7730 Ergonomics of the thermal environment–Analytical determination and interpretation of thermal comfort
using calculation of the PMV and PPD indices and local thermal comfort criteria; 2005
65
P.O. Fanger, J. Toftrum, Extension of the PMV model to non-air-conditioned buildings in warm climates, Energy and
Buildings 2002; 34:533-53
Olesen B: Thermal comfort. Technical Review 1982, 2:3–37
European Committee for Standardization (CEN). EN 15251 indoor environmental input parameters for design and
assessment of energy performance of buildings – addressing indoor air quality, thermal environment, lighting and
acoustics. Brussels; 2007.
de Freitas, C. and Grigorieva, E. (2014). A comprehensive catalogue and classification of human thermal climate
indices. International Journal of Biometeorology, 59(1), pp.109-120.
Parsons, K. (2006). Heat Stress Standard ISO 7243 and its Global Application. Industrial Health, 44(3), pp.368-
379.Blazejczyk K, Epstein Y, Jendritzky G, Staiger H, Tinz B. Comparison of UTCI to selected thermal indices. Int J
Biometeorol 2012;56:515–35.
Winslow, C., Herrington, L. and Gagge, A. (1937). PHYSIOLOGICAL REACTIONS OF THE HUMAN BODY TO
VARYING ENVIRONMENTAL TEMPERATURES. American Journal of Physiology-Legacy Content, 120(1),
pp.1-22.
Gagge, A P, Stolwijk, J A J & Nishi, Y (1971): An effective temperature scale based on a simple model of human
physiological regulatory response. ASHRAE Trans. 77(pt.1): 247-262
Lee DHK. Proprioclimates of man and domestic animals. In: Climatology: Reviews of Research. Paris (France):
UNESCO; 1958. p. 102e25.
R. Bhattacharya, G. Biswas & 3A. Bhattacharya (2014): Degree of Discomfort Over Assam Using Meteorological
Data. International Journal of Electronics & Communication Technology.
U.S. Department of Energy (2011): Building energy data book
Mirinejad, H., Welch, K. C., & Spicer, L. (2012). A review of intelligent control techniques in HVAC systems. 2012
IEEE Energytech, (May 2012), 1–5. https://doi.org/10.1109/EnergyTech.2012.6304679
G. Zhang et al., Real-life Applications with Membrane Computing, Emergence, Complexity and Computation 25,
DOI 10.1007/978-3-319-55989-6_2
Schroeders, U., Wilhelm, O. and Olaru, G. (2016). Meta-Heuristics in Short Scale Construction: Ant Colony
Optimization and Genetic Algorithm. PLOS ONE, 11(11), p.e0167110.
Agatonovic-Kustrin, S. and Beresford, R. (2000). Basic concepts of artificial neural network (ANN) modeling and its
application in pharmaceutical research. Journal of Pharmaceutical and Biomedical Analysis, 22(5), pp.717-727.
R. Alcal'a, J.M. Ben'ttez, J Casillas, O. Cordo'n, and R Pe'rez, "Fuzzy control of HV AC systems optimized by genetic
algorithms”, Applied Intelligence, vol. 18, pp. 155-177,2003.
Guo, P.Y., Z.H. Guang and Z. Bien, 1998. A simple fuzzy adaptive control method and application in HVAC.
Proceeding of the IEEE International Conference on Fuzzy System, IEEE World Congress on Computational
Intelligence, May 4-9, Anchorage, Alaska, USA., pp: 528-532.
Data Mining with Decision Trees: Theory and Applications 2015 3 Edited by Lior Rokach and Oded Maimon. World
Scientific Publishing 2015. 39(3), pp.437-438.
Choi, Joon-Ho, Azizan, Aziz, and Vivian, Loftness. 2010. “Investigation on the impacts of different genders and ages
on satisfaction with thermal environments in office buildings." Building and Environment 45 (6), 1529-1535.
66
Ellis, F. (1953). Thermal comfort in warm and humid atmospheres: Observations on groups and individuals in
Singapore. Journal of Hygiene, 51(03), pp.386-404.
Bedford, T. (1948). Basic Principles of Ventilation and Heating.
Abstract (if available)
Abstract
Nowadays, there are various choices of building control strategies for improving thermal comfort while saving energy. Using real time simulation to control building heating ventilation, air conditioning (HVAC) system has been studied in the building science realm for many years. However, the following two aspects of the building control have not been studied thoroughly in research. One is combining optimization methods to control the set-points on HVAC to minimize thermal stress for multi-occupant office and the energy consumption, and the other is generating trade-off lines between thermal stress and energy use. This research integrates Matlab, Python and EnergyPlus with a Data-Driven Rule-Based Control (DRBC) algorithm method to predict optimal building control strategies. ❧ This paper proposes an overall thermal comfort model by undergoing experiments. The human thermal comfort experiments were conducted to collect occupants’ thermal preference by surveying their thermal comfort feedback under different environmental conditions in an environment chamber. The thermal comfort and energy saving model is proposed to be the objective set-point selecting. Moreover, sensors are implemented in the chamber to collect individual indoor thermal condition data, such as indoor air temperature, relative humidity, ventilation rate, mean radiant temperature. DRBC is introduced to find the optimum HVAC dynamic control set points, ventilation mode, temperature and wind speed, of the diffuser. In addition, the final decision will be affected by both the energy saving and the thermal comfort of set points optimization process. ❧ The results represent that the use of Energyplus coupled with DRBC for the application of HVAC set-point control. Besides, albeit the selected set-point combination of HVAC control strategies is not guaranteed to be the best, it will be at least satisfied the most occupants in an area. Finally, DRBC set-point is possible to improve the efficiency over ASHERAE standard while achieving occupants’ thermal comfort.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Exploration for the prediction of thermal comfort & sensation with application of building HVAC automation
PDF
Multi-occupancy environmental control for smart connected communities
PDF
Enhancing thermal comfort: data-driven approach to control air temperature based on facial skin temperature
PDF
Real-time simulation-based feedback on carbon impacts for user-engaged temperature management
PDF
Learning personal thermal comfort and integrating personal comfort requirements into HVAC system control loop
PDF
Developing a data-driven model of overall thermal sensation based on the use of human physiological information in a built environment
PDF
Enhancing thermal comfort: air temperature control based on human facial skin temperature
PDF
Exploring participatory sensing and the Internet of things to evaluate temperature setpoint policy and potential of overheating/overcooling of spaces on the USC campus
PDF
Occupant-aware energy management: energy saving and comfort outcomes achievable through application of cooling setpoint adjustments
PDF
Natural ventilation in tall buildings: development of design guidelines based on climate and building height
PDF
Human–building integration: machine learning–based and occupant eye pupil size–driven lighting control as an applicable visual comfort tool in the office environment
PDF
Nursing comfort study: are hospitals performing for nurses? IEQ analysis of working conditions inside hospitals
PDF
A simplified building energy simulation tool: material and environmental properties effects on HVAC performance
PDF
Building energy performance estimation approach: facade visual information-driven benchmark performance model
PDF
Human-building integration based on biometric signal analysis: investigation of the relationships between human comfort and IEQ in a multi-occupancy condition
PDF
Smart buildings: employing modern technology to create an integrated, data-driven, intelligent, self-optimizing, human-centered, building automation system
PDF
Impact of occupants in building performance: extracting information from building data
PDF
Economizer performance and verification: effect of human behavior on economizer efficacy and thermal comfort in southern California
PDF
Impacts of building performance on occupants' work productivity: a post occupancy evaluation study
PDF
Visualizing thermal data in a building information model
Asset Metadata
Creator
Chen, Li-Chen
(author)
Core Title
Developing environmental controls using a data-driven approach for enhancing environmental comfort and energy performance
School
School of Architecture
Degree
Master of Building Science
Degree Program
Building Science
Publication Date
08/15/2019
Defense Date
08/15/2018
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
building energy,data-driven control,HVAC,indoor thermal control,OAI-PMH Harvest,thermal comfort
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Choi, Joon-Ho (
committee chair
), Narayanan, Shrikanth (
committee member
), Schiler, Marc (
committee member
)
Creator Email
b99602054@gmail.com,lichenc@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c89-71923
Unique identifier
UC11670353
Identifier
etd-ChenLiChen-6766.pdf (filename),usctheses-c89-71923 (legacy record id)
Legacy Identifier
etd-ChenLiChen-6766.pdf
Dmrecord
71923
Document Type
Thesis
Format
application/pdf (imt)
Rights
Chen, Li-Chen
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
building energy
data-driven control
HVAC
indoor thermal control
thermal comfort