Renewable and Sustainable Energy Reviews: Zakia Afroz, GM Sha Fiullah, Tania Urmee, Gary Higgins

Renewable and Sustainable Energy Reviews xxx (xxxx) xxx–xxx
Contents lists available at ScienceDirect
Renewable and Sustainable Energy Reviews

journal homepage: www.elsevier.com/locate/rser
Modeling techniques used in building HVAC control systems: A review

⁎
Zakia Afroza, , GM Shafiullaha,b, Tania Urmeea,b, Gary Higginsc
a
School of Engineering and Information Technology, Murdoch University, 90 South Street, Western Australia, 6150, Australia
b
Lecturer of Electrical Engineering, Energy and Physics, School of Engineering and Information Technology, Murdoch University, Australia
c
Energy Management Facility, Murdoch University, 90 South Street, Western Australia, 6150, Australia
A R T I C L E I N F O A B S T R A C T
Keywords: The appropriate application of advanced control strategies in Heating, Ventilation, and Air-conditioning (HVAC)
Modeling technique systems is key to improving the energy efficiency of buildings. Significant advances have been made in the past
Building decades on model development to provide better control over the energy consumption of system components
HVAC while simultaneously ensuring a satisfactory indoor environment in terms of thermal comfort and indoor air
Control system
quality. Yet it is an ongoing challenge to select and implement the best-suited modeling technique for improving
Energy consumption
the control strategy of HVAC systems. For the development of modeling research it is important that the building
Thermal comfort
Indoor air quality research community is informed about the role, application, merits, shortcomings and outcomes of different
modeling techniques used in HVAC systems. Even though several review articles have been published on
modeling techniques, the weaknesses and strengths of these modeling techniques, along with performances of
developed models associated with research studies, have rarely been identified. This study presents a critical
review of current modeling techniques used in HVAC systems regarding their applicability and ease of accep-
tance in practice and summarizes the strengths, weaknesses, applications and performance of these modeling
techniques. Additionally, the performance and outcome of some of the developed models used in real world
HVAC systems have been discussed. From the extensive critical review it is evident that almost every model has a
major/minor shortcoming generated from assumptions, unmeasured disturbances or uncertainties in some
system properties. This review aims at highlighting the shortcomings of existing application-based research on
HVAC systems, and accordingly, recommendations are presented to improve the performance of building HVAC
systems.
1. Introduction control algorithms [6–17]. Improving the system control algorithm,

mandates accurate modeling of the system and implementation of best-
HVAC is the primary energy consuming sector in a building [1]. suited optimization techniques [18].
Therefore, improving the efficiency of HVAC systems contributes to Modeling techniques are increasingly getting sophisticated as ad-
greater energy savings within the building [2]. HVAC energy demand is vances in HVAC control system. HVAC, as practiced today, not only
directly related to the indoor temperature set-point, air infiltration, takes into account the loss or gain of heat through the windows, walls,
window type, window-wall ratio, and internal loads [3]. Additional floors, ceilings and roofs of buildings and the supply of fresh air needed
influences include building type and climate [3]. Research has identi- on account of physical and process contamination in addition to reg-
fied that appropriate selection and operation of HVAC systems can ulatory requirements, but also addresses control of the temperature,
provide energy saving by 25% while maintaining satisfactory indoor humidity and cleanness of the air and above all, its movement [19].
environment [4]. However, poor design and inefficient operation of Considering all of these discrete, non-linear and highly constrained
HVAC system can result in excessive energy consumption. characteristics and parameters of HVAC systems, it is a challenging task
The efficient operation of an HVAC system largely depends on its to develop an accurate and effective model for these systems that ac-
control system and optimization parameters. It is far more sustainable curately represents reality. For the development of modeling research it
and cost effective to improve the control algorithms to achieve higher is necessary that the building research and management community
efficiency than replacing HVAC equipment with more efficient modern become informed about the application, role, strengths and weaknesses
technologies [5]. A good number of recent articles have given emphasis of the various modeling techniques associated with research studies and
on improving the energy efficiency in buildings through advanced how the developed models perform in real world situations. A review
⁎
Corresponding author.
E-mail address: [email protected] (Z. Afroz).
https://doi.org/10.1016/j.rser.2017.10.044
Received 24 February 2017; Received in revised form 29 June 2017; Accepted 26 October 2017
1364-0321/ © 2017 Elsevier Ltd. All rights reserved.
Please cite this article as: Afroz, Z., Renewable and Sustainable Energy Reviews (2017), https://doi.org/10.1016/j.rser.2017.10.044
Z. Afroz et al. Renewable and Sustainable Energy Reviews xxx (xxxx) xxx–xxx
Nomenclature Te Temperature of the refrigerant at evaporator inlet (°C)

Th Supply air temperature (in humidifier) (°C)
cov Covariance Tm Temperature of the air out of the mixing box (°C)
fsa Volumetric flow rate of the supply air To Outside temperature (°C)
fsw Water flow rate Tr Temperature of the recirculated air (°C)
h(t) Rate of humid air produced in the humidifier TR Temperature of the roof (°C)
hs Enthalpy of saturated air at the temperature of the wetted Tsi Temperature of supply air (to the humidifier) (°C)
surface (kJ/kg) Tsa Supply air temperature (°C)
h1 Enthalpy of the refrigerant at evaporator outlet/com- Tw1 Temperature of the wall (East, West) (°C)
pressor inlet (kJ/kg) Tw2 Temperature of the wall (South, North) (°C)
h2 Enthalpy of the refrigerant at compressor outlet/con- Twb,i Wet-bulb temperature of ambient air (°C)
denser inlet (kJ/kg) Twi Supply water temperature (°C)
h5 Enthalpy of the refrigerant condenser outlet/expansion Two Return water temperature (°C)
valve inlet (kJ/kg) Tz Temperature of the zone (°C)
h6 Enthalpy of the refrigerant expansion valve exit/eva- Twe Temperature of the evaporator wall (°C)
porator inlet (kJ/kg) Twc Temperature of the condenser wall (°C)
mo Mass flow rate of the outdoor air Uw1 Overall heat transfer coefficient of (East, West) walls
mr Mass flow rate of the recirculated air Uw2 Overall heat transfer coefficient of (South, North) walls
ṁ r Mass flow rate of the refrigerant UR Overall heat transfer coefficient of the roof
mm Total mass flow rate of the mixing air Vah Volume of the air handling unit
ṁ w Mass flow rate of condenser water (kg/s) Vh Volume of the humidifier
q−1 Back shift operator Vz Volume of the zone
q(t) Heat gains from occupants, and light (W) Wco Humidity ratio of the air out from the coil (kg/kg dry air)
Aei Area of the evaporator inlet (m2) Wh Supply air humidity ratio (in humidifier) in kg/kg(dry air)
Aci Area of the condenser inlet (m2) Wm Humidity ratio of the air out the mixing box (kg/kg dry
Aeo Area of the evaporator outlet (m2) air)
Aco Area of the condenser outlet (m2) Wsa Humidity ratio of the supply air in kg/kg (dry air)
AR Area of the roof (m2) Wsi Humidity ratio of the supply air (to the humidifier) in kg/
Aw1 Area of the wall (East, West) (m2) kg (dry air)
Aw2 Area of the wall (South, North) (m2) Wz Humidity ratio of the zone in kg/kg (dry air)
Cah Overall thermal capacitance of the air handling unit ( kJ/ ∆y (∞) Amplitude of step chage in process output
°C) Δu (∞) Amplitude of step chage in process input
Ch Overall thermal capacitance of the humidifier (kJ/°C)
CR Overall thermal capacitance of the roof (kJ/°C) Greek letters
Cw1 Overall thermal capacitance of the wall (East, West) (kJ/
°C) α, β, γ, δ, θ Polynomials
Cw2 Overall thermal capacitance of the wall (South, North) τ Time constant
(kJ/°C) σ Standard deviations
Cz Overall thermal capacitance of the zone (kJ/°C) ε Effectiveness of heat exchanger (dimensionless)
Cpw Specific heat of water=4.1868 kJ/kg °C ρa Density of air=1.25 kg/m3
Cpa Specific heat of air=1.005 kJ/kg °C ρw Density of water=998 kg/m3
Cpa,e Specific heat of equivalent ideal gas (kJ/kg °C) ω Ratio of the minimum heat capacity flow rate to the
K Static gain maximum heat capacity flow rate (dimensionless)
L Dead time αh (UA)h overall transmittance area factor of the humidifier
MeM Merkel number (dimensionless) αci Heat transfer coefficient of refrigerant entering condenser
NTU Number of transfer units (dimensionless) (W/m2 K)
P(t) Evaporation rate of the occupants αco Heat transfer coefficient of refrigerant leaving condenser
Qċ Heat Transfer rate of condenser (W/m2 K)
Qė Heat Transfer rate of evaporator αei Heat transfer coefficient of refrigerant entering evaporator
Q̇rej Heat rejection rate of the cooling tower (W) (W/m2 K)
Tc Temperature of the refrigerant at condenser inlet (°C) αeo Heat transfer coefficient of refrigerant leaving evaporator
Tco Outlet temperature of the air from the coil (°C) (W/m2 K)
Tcws Condenser water supply temperature (°C)
Tcwr Condenser water return temperature (°C)
article addressing these issues can provide a deep insight into the ex- systems. A brief review of three major modeling techniques for HVAC
isting modeling techniques used in HVAC system. system was reported by Homod [21] even though their review did not
Several review articles have been published so far on modeling cover the subdivision of those three major modeling techniques. Okochi
techniques [20–23], control strategies [24–27] and optimization and Yao [22] presented a review of variable air volume (VAV) systems’
methods [27] of HVAC systems. Afram and Janabi-Sharifi [20] re- modeling and simulations, controls, airflow properties, common faults‫׳‬
viewed the modeling methods used in heating, ventilation, and air diagnosis and control, energy consumption analysis, and current ap-
conditioning (HVAC) systems. In another research paper, Afram and plications. A review of major modeling methodologies used to model
Janabi-Sharifi [26] reported control methods, with an emphasis on the the energy systems of buildings was presented by Harish and Kumar
theory and applications of model predictive control (MPC) for HVAC [23]. Dounis and Caraiscos [24] reviewed the control systems used in
2
building energy and comfort management. A comprehensive review of

existing supervisory and optimal control strategies for HVAC system
was reported by Wang and Ma [25]. Shaikh et al. [27] reviewed the
control systems, optimization methods and simulation tools used in
building energy and comfort management and analyzed the energy and
comfort related trends.
Also, Trčka and Hensen [28] reported a review of existing tools for
HVAC system design and analysis, modeling approaches, and simula-
tion techniques. Subbaram Naidu and Rieger [29,30] presented an
overview of advanced control strategies for HVAC systems including
hard, soft and fusion control. A comprehensive review of intelligent
control techniques used in HVAC systems was provided by Mirinejad
et al. [31].
There is still a need to understand the strengths and weaknesses of
different modeling methods used in HVAC system to progress in re- Fig. 1. Schematic of typical chilled-water ventilation and air-conditioning system for
search on modeling of HVAC system components and operating con- commercial buildings (Reprinted from [34] with permission from Elsevier).
ditions. The aim of this paper is twofold:
comprising three main components: air handling unit, chiller and
i. Categorize the role of different modeling techniques in improving cooling tower. The system also consists of several sensors and con-
the function of HVAC control system; functional improvement in- trollers for regulating the controllable variables such as zone tem-
cludes reducing energy consumption, maintaining thermal comfort perature, supply air temperature, supply air fan speed, duct static
and improving indoor air quality. pressure, and chilled water temperature at their set-points [33]. Be-
ii. Indicate the need for further research efforts cause of having numerous mechanical, hydraulic and electrical com-
ponents, the overall dynamics of HVAC plants are highly nonlinear. The
Therefore, this study presents a critical survey of different modeling interaction between the temperature and humidity control loops is re-
methods used in HVAC system associated with various research studies. latively complex, and substantial constraints are imposed by the non-
This study also provides a deep insight into the applications, tasks, and ideal behavior of actuators such as dampers and valves [18]. Therefore,
outcomes of some of the developed models where these modeling the modeling process of the HVAC system leads to dynamic, nonlinear,
techniques have been implemented in the real world. Through ex- and very high-order models because of the physical properties such as
tensive review, this study will answer the following research questions: high-thermal-inertia, real lag time, uncertain disturbance factors, etc. of
the system [21].
a. What are the roles, application, strengths and weaknesses of dif- The most challenging and important part of the model development
ferent modeling methods used in HVAC system control system? for a particular application is the process of identifying the model order
b. What were the goals and outcomes of the developed models asso- and the optimum parameters [35]. The adjustment of parameters of a
ciated with research studies and how these models performed in model used to denote a system is called system identification [36].
practice? and System identification, is often a necessary tool for the design, com-
c. What are the major performance criteria of these modeling techni- missioning, operation, control, optimization and diagnostic processes of
ques and how these performance criteria can help in selecting the HVAC systems in order to describe the behavior and/or dynamics of air
suitable modeling technique based on field of application? conditioning systems [37]. A Properly identified model can provide
good results, higher accuracy and at the same time may require
This study will act as a reference for the building research com- minimum calculation time. Agbi et al. [38] reported that parameter
munity and guideline for the energy management personnel to select identifiability is an important factor to system identification accuracy
the best-suited modeling technique on the way to improve the perfor- and efficiency, and it is influenced by the input data, excitation signals,
mance of the HVAC system. and system model structure. On the other hand, Li and Wen [39] in
This paper is organized as follows: Section 2 addresses the necessity their studies mentioned that like the model structure selection, the
of accurate modeling in HVAC system. The step-by-step methodological system order and Hankel matrix1 size are the two vital factors for en-
approach that is followed during review has been specified in Section 3. suring model accuracy and high calculation speed in system identifi-
In Section 4 the modeling techniques used in HVAC systems have been cation. They conducted parametric testing in their studies to determine
introduced and classified. These modeling techniques have been criti- the system order and the Hankel matrix size. It was found from their
cally reviewed in Section 5 to identify the features, strengths and model validation that 18th order subspace model performed the best in
weaknesses of these techniques. This Section also presents a review of terms of simplicity and high precision. However, even though para-
application-based research studies of modeling implementations. metric testing methods were found to be useful for determining the
Section 6 summarizes the outcome of the review. This section also system order and Hankel matrix rank, there is still a lack of systematic
provides an insight into the direction for developing models in the fu- approach to the model structure selection, model order determination
ture. Lastly, a conclusion is drawn from the entire study in Section 7. and optimum parameters identification [39]. Most of the existing stu-
dies just applied trial and error method to determine the model struc-
2. Necessity of choosing appropriate modeling processes in HVAC ture and order. Despite the similarity of HVAC control to other types of
control system process control, certain features such as nonlinear dynamics, time-
varying system dynamics and set-points, time-varying disturbances,
Modeling of HVAC system is essential for appropriate analysis and poor data, complex interaction between the temperature and humidity
improvement of its control system. However, HVAC systems possess a control loops and at times conflicting control loops and lack of
complex structure comprising heat and mass transfer equipment such as
the chiller, boiler, heating/cooling coils, thermal storage systems, air-
handling equipment, air distribution system and liquid distribution 1
A Hankel matrix is a matrix that is symmetric and constant across the anti-diagonals,
systems [32]. Fig. 1 shows the schematic of a conventional chilled- and has elements h(i,j) = p(i+j-1), where vector p = [c r(2:end)]. This p completely
water ventilation and air-conditioning system for commercial buildings determines the Hankel matrix.
3
supervisory control (in many buildings) has made HVAC system control a. Classification of HVAC Modeling Techniques based on literature and
distinctive and challenging [40]. Therefore, while developing a model b. Extensive study on the classified HVAC Modeling Techniques asso-
for HVAC system or its components, attention should be given in de- ciated with existing relevant research studies
termining the model order and parameter identification so that the
ultimate control algorithm holds the ability to deal with disturbances, After completion of the review process the findings have been dis-
constraints, and uncertainties, to handle time-varying system dynamics cussed and summarized. In the discussion and summary section a
and slow-moving processes with time delays and to cope with a broad comparison of modeling techniques is presented based on weaknesses
range of operating conditions. and strengths of different modeling techniques and their performance
criteria. Lastly, recommendations are presented to progress in research
on modeling of HVAC system components and operating conditions.
3. Methodology
4. Classification of HVAC modeling techniques

The review was conducted by the authors based on exploring/
searching:
HVAC modeling techniques can be divided into three categories:
physics-based (or white box/mathematical/forward), data-driven (or
I. publication databases for peer-reviewed journal and conference ar-
black box/empirical/inverse), and gray box (or hybrid). These models
ticles, dissertations, and books, and
can be linear [41,42] or nonlinear [43,44], static [45–53] or dynamic
II. technical reports, and documentations on the application of HVAC
[43,54–59], explicit or implicit, discrete or continuous, deterministic
modeling techniques used for several decades.
[60] or probabilistic, deductive, inductive or floating. Static/steady
state models are those in which parameters are constant or do not vary
The literature search was carried out using related keywords,
with time while dynamic/unsteady state/transient models are those in
identified through an iterative process of study, brainstorming and
which parameters change with time. Dynamic modeling deals with
combination of both. Three different categories of keywords were used:
various problems arising in transient operations that are crucial for
system start-up, shut-down, and reaction to disturbances [61]. Dis-
a. Modeling techniques (e.g., physics based, data-driven and hybrid)
turbances in HVAC systems can take place due to heating and cooling
used for HVAC system,
load changes, human interactions, and/or control actions [62].
b. Application of major modeling techniques and their sub-sections in
According to this grouping, most of the physics-based techniques
HVAC system (e.g., Modeling various HVAC system dynamics,
fall under the deductive models whereas data-driven techniques fall
control of indoor temperature, thermal comfort level, heating/
under the inductive models. Gray-box models can be classified under
cooling load, optimization of HVAC energy consumption, fault de-
both the inductive and deductive models. Both physics-based and data-
tection, and diagnosis of HVAC systems, etc.), and
driven techniques can result in linear/nonlinear, static/dynamic, and
c. Generic terms (e.g., building energy conservation, HVAC system
explicit or implicit models. Physics-based techniques generally lead to
optimization, indoor air quality, indoor thermal comfort) to ensure
continuous and deterministic models whereas data-driven techniques
the breadth and depth of coverage.
commonly result in discrete and deterministic or stochastic models
Search terms were combined with Boolean operators, “OR” and
[63,64].
“AND” to cover extensive possible combinations in search engines.
The following key websites were used to identify potential sources:
● IEEE Xplore (http://ieeexplore.ieee.org/Xplore/home.jsp), 4.1. Physics-based (white box/mathematical/forward) model
● Science Direct (http://www.sciencedirect.com),
● Scopus (http://www.scopus.com), Physics-based models are developed based on fundamental laws of
● Google Scholar (https://scholar.google.com.au) and energy - mass balance, heat transfer, momentum, and flow balance,
● Murdoch Library (http://library.murdoch.edu.au) from where a set of mathematical equations can be derived and solved.
These models can be either distributed or lumped parameter type.
Cited articles were checked and relevant cited articles were in- Based on ease of use, the lumped parameter method has shown superior
cluded in the review. performance over distributed parameter type [21]. Several specific as-
The review was conducted mainly in the following areas: sumptions are the basis of physics-based models. This type of models is
Sub-Space State Fig. 2. Classification of data-driven models used in

First/ Second Order Over-damped
Space Identification Process with Dead Time HVAC systems (adopted from [20]).
(4SID)
State-Space Frequency
Model Domain Model Artificial Neural Network (ANN)
Thin Plate Spline
(TPS) Approximation Data Mining Support Vector Machine (SVM)
Algorithm
Geometric Fuzzy Adaptive Network (FAN)
Topological Case Base
Modeling (TCBM) Model
Takagi-Sugeno (T-S) Fuzzy
Probability Density Case-Based Fuzzy Logic
Function (PDF) Reasoning Data-driven Modeling Model
Adaptive Network based Fuzzy
Approximation Types Inference System (ANFIS)
Stochastic
Model Linear and Polynomial Time
Just In Time (JIT) Series Regression
Instantaneous Statistical
Model Model Auto Regression exogeneous
(ARX)
Auto Regression Moving Average
exogeneous (ARMAX)
Auto Regression Integrated
Moving Average (ARIMA)
4
primarily used at the design stage where it is necessary to predict and 5.1.1. Chiller model
analyze the performance of HVAC system components through simu- The chiller is a primary system component of an HVAC system that
lation. Examples of physics-based model include: chiller model rejects heat from a liquid through a vapor compression cycle or an
[60,65–71], cooling tower model, zone model [54,72–75], AHU model, absorption cooling cycle [91]. Chillers, chilled water pumps, and mo-
mixing box model [54,72,74], splitting box model [72], heating/ tors altogether consume almost half of the total energy used in com-
cooling coil model [54,72,73,75–80], humidifier model [54], fan or mercial and institutional buildings [92,93]. Chillers, being a major
pump model [54,72,73,75], duct model [54], sensor model [54], energy consuming component of HVAC systems, hold vital significance
damper or valve model [72], etc. in developing simulation models for effective control and optimal op-
Physics-based dynamic HVAC system models are commonly devel- erations of HVAC systems. A chiller comprises four elementary com-
oped for the slow moving temperature and humidity processes (e.g., ponents: evaporator, condenser, compressor, and expansion valve. All
zone temperature dynamics, zone humidity dynamics, heating/cooling these components can be modeled separately.
coil dynamics, etc.), while static models are implemented for the fast The chiller power consumption differs based on the water flow rate,
moving dynamics of the system [e.g. mixed air temperature and carbon the heat capacity of water, the temperature difference between in-
dioxide (CO2) concentration in mixing box, and flow rate of air and coming and outgoing water in a chilled water loop and the Coefficient
water through damper and valve respectively] and energy consumption of Performance (COP) of the chiller [94]. COP of the chiller varies based
(fan or pump energy consumption) [20]. on the load on the chiller. Additionally the efficiency of a cooling tower
is directly related to the enthalpy of the air to which heat is being re-
jected.
4.2. Data-driven (black box/inverse) model
As dynamic models can better represent the transient behavior of
the chiller system, a significant amount of research has been conducted
Data-driven models are developed through a process of collecting
on the dynamic modeling of chiller system. Browne and Bansal [95]
the system performance data from real practice and then establishing a
presented a transient simulation model for predicting the dynamic
relationship between the input and output variables using the mathe-
performance of vapor-compression liquid chillers over a wide range of
matical techniques [e.g., statistical regression or artificial neural net-
operating conditions. In one of the chillers they utilized a regression
work (ANN)] [32]. The ability of data-driven approaches in modeling
model for the compressor and also had to apply empiricism for the
HVAC system has been established in various research studies
evaporator tube wall mass to predict the start-up process for that par-
[9,56,81–89]. This type of modeling is suitable for existing HVAC
ticular chiller adequately.
system performance improvement where sufficient training data are
The main assumptions used in this study to develop the model are:
available.
4.3. Classification of data driven models

• The mass flow rate of the refrigerant is equal to the mass flow rate
through the compressor and keeps same throughout the system.
Data driven (black box or inverse) modeling technique comprises

• The temperature of the walls does not vary through the cross-section
or along the length of the tubes.
nine different types of models. A complete classification of data-driven
models used in HVAC system is illustrated in Fig. 2.
• The refrigerant properties within each component are homogeneous
with pressure drops being neglected.
• The expansion process through the EEV/orifice plate is isenthalpic.
4.4. Gray-box (hybrid) model
Assuming that the refrigerant is in a quasi-steady state, the eva-
Gray box models can be regarded as the combination of white box porator load and the energy balance equations used by Browne and
and black box models [21]. The basic structure of the model is formed Bansal [95] are given by:
from physics-based methods while the model parameters are de-
Qė = α ei Aei (Two − Twe ) (1)
termined by using the parameter estimation algorithms on the mea-
sured data of the system. In fact, there are some physical processes of ṁ r (h1 − h6) = α eo Aeo (Twe − Te ) (2)
the HVAC system that are less clearly defined by the thermodynamic
equations, and at the same time, certain information or adequate Similarly, for the condenser the load and energy balance equations
training data are lacking. In this case, gray box or hybrid model is ef- are given by:
fective. This approach is especially beneficial for control applications
Qċ = α ci Aci (Twc − Tcwr ) (3)
when the model is expressed in a suitable form such as transfer function
or state space [90]. ṁ r (h2 − h5) = α co Aco (Tc − Twc ) (4)
The prediction result of this study [95] was found to be within ±

5. A Review on HVAC modeling techniques
10%, even though for one of the chillers a degree of empiricism was
applied and a simple cooling-tower model was developed based on the
HVAC system dynamics modeling has been used for decades in
dynamics of the water sump only. Many other aspects of system tran-
design, control, fault detection and diagnosis, and improving the per-
sients were also neglected while developing the model. Therefore, the
formance of HVAC systems. This section provides a critical review of all
model was not fully capable of predicting the modulating nature of the
modeling techniques in terms of their strengths and weaknesses. This
chiller.
review also pinpoints the methodological approach, application, out-
Wang et al. [60] presented dynamic models for single-stage and
come and shortcoming of some of the developed models used in HVAC
two-stage centrifugal chillers on the basis of fundamental mechanistic/
system associated with research studies.
deterministic principles of chillers. The simulated dynamics of chillers
was used by this study to test the control performance of an HVAC
5.1. Physics-based model system. The model was developed considering the compressor poly-
tropic work, hydrodynamic losses and mechanical and electrical losses.
Physics-based modeling techniques are applied to HVAC system Subsequently, it was validated by comparing the predicted chiller
components and its sub-systems. These are discussed in the following power consumption and coefficient of performance with the measured
subsections. data. This model was found useful to identify the effect of chiller
5
dynamics on the HVAC system although it was not suitable to evaluate During modeling of the zone, Tashtoush et al. [54] assumed that:
the dynamic performance of chiller operation. Also, some additional
assumptions were used during model development along with basic • The air in the zone is fully mixed, and zone temperature distribution
assumptions of the thermodynamic cycle which ultimately affected the is uniform.
prediction accuracy. • The effect of North and South wall on the zone temperature is same.
Based on first principles of thermodynamics Bendapudi et al. [66] Also, the effect of East and West wall on the zone temperature is
proposed a dynamic model for a centrifugal chiller. The study found same.
that second-order integration algorithm, especially the modified Euler • The ground/floor has no effect on the zone temperature.
method, in conjunction with a suitable integration step-size provides • The density of the air is constant and is not influenced by changing
the fastest execution speed as compared to a first-order or fourth-order the temperature and humidity ratio of the zone.
method. This proposed model was found to predict the steady state • No pressure losses across the zone and in the mixing section.
performance under a wide variety of conditions as well as the time
constants of the transients. However, the shell-and-tube heat ex- Under the above assumptions, energy and mass balance governing
changers were modeled based on finite volume approach with the re- equations of the zone expressed by Tashtoush et al. [54] are given by:
frigerant pressure drop neglected [61]. In a later work, Bendapudi et al.
dTz
[96] compared two common modeling approaches [i.e., finite volume Cz = fsa ρa Cpa (Tsa − Tz) + 2Uw1 Aw1 (Tw1 − Tz) + UR AR (TR − Tz)
dt
(FV) and moving boundary (MB) methods] for heat exchanger models
where the refrigerant pressure drop was again neglected to simplify the + 2Uw2 Aw2 (Tw2 − Tz) + q (t ) (5)
calculation. According to their research findings, the MB method is Where q (t ) stands for internal heat gain from the occupancy, lighting
much faster than the FV method although both of them provide ap- fixtures, other devices and can be expressed as
proximately the same accuracy. However, the FV method being capable
of dealing with the start-up transient shows more robustness than the q (t ) = qp + ql (6)
MB method. It was concluded that neither approach could predict the
Eq. (18) states that the rate of change of energy in the zone is equal
refrigerant charge accurately due to lack of appropriate void-fraction
to the difference between the energy transferred to the zone by either
models.
conduction or convection and the energy removed from the zone.
Lei and Zaheeruddin [69] proposed a lumped parameter dynamic
model of a water-chiller refrigeration system based on mass and energy dTw1
Cw1 = Uw1 Aw1 (Tz − Tw1) + Uw1 Aw1 (To − Tw1)
balance principles. A simulation study was performed to examine the dt (7)
transient behavior of the dynamic model subject to variations in com-
pressor operational frequency and a thermodynamic expansion valve dTw2
Cw2 = Uw2 Aw2 (Tz − Tw2) + Uw2 Aw2 (To − Tw2)
dt (8)
opening fraction. The study identified the optimum valve position at a
given operation frequency to work within a safe operating mode to dTr
avoid water freeze up in the secondary system. Even though the studies CR = UR AR (Tz − TR ) + UR AR (To − TR )
dt (9)
incorporated the refrigerant thermal capacity with superheat section of
evaporator, certain assumptions (i.e. no effect of the compressor shell Eqs. (20), (21) and (22) states that the rate of change of energy in
on system performance, compressor reaches operating speed in- the walls is equal to the energy transferred to the walls due to the
stantaneously, no pressure drop and longitudinal heat transfer within temperature difference between indoor and outdoor air.
the heat exchangers, no refrigerant mass accumulation in the superheat dWz P (t )
section) have made the simulation model simple. Therefore, there is a Vz = fsa (Wsa − Wz) +
dt ρa (10)
possibility that this model will differ largely in performance from real
practice. Similarly, Eq. (23) states that the rate of change of moisture content
A dynamic model for single-effect LiBr/water absorption chillers in the zone is equal to the difference between the vapor added or re-
based on external and internal steady-state enthalpy balances for each moved from the zone.
main component was presented by Kohlenbach and Ziegler [65,68]. A model of the zone has to emulate the dynamic thermal processes
This model simulated the reaction of the absorption chiller to a change within the zone as well as the interaction with the environment because
of external conditions. Assumptions were made to reflect the ideal of the continuous variation of the external and internal environment of
condition. Also, a simpler approach was applied considering the most the zone [97]. Therefore, modeling this time-variant system is a chal-
important physical properties only that avoided a detailed enthalpy lenging task especially, when the number of zones becomes multiple. In
calculation for each state point. single-zone buildings, the set points of thermal comfort and indoor air
Zhang et al. [70] developed a dynamic model for an air-cooled quality variables are the same in all rooms, while in multi-zone build-
screw chiller consisting of an economized screw compressor for con- ings, the set points vary for different zones. It is easier to design a
troller design purposes. Comparisons made between simulations and controller for a single-zone building where many factors such as zone
experiments specified that the model could sensibly simulate the tran- overlapping, variation in heating/cooling load, solar radiation or
sients of the chiller. However, the study overlooked the dynamics of the change in thermal effects among zones are not reflected. In the case of
compressor by assuming the compression process to be polytropic and multi-zone building HVAC systems, such thermal coupling cannot be
static. neglected and must be modeled accurately for the effective control of
zone temperature, humidity, and air quality [100]. This strategy results
5.1.2. Zone model in more complex multi-input-multi-output (MIMO) controllers.
An HVAC zone can be defined as a cluster of adjacent offices and/ or Often the zone model is simplified to linear single zone ignoring
spaces often covered by a common air-handling unit (AHU) or air time varying characteristics of parameters. For instance, Ghiaus and
terminal device [97]. The temperature in a zone is maintained by Hazyuk [101] assumed that the thermal model of the building is linear
adding or removing the heat into or from the space to balance the in- and that the model of the building, the weather conditions, and occu-
ternal and external heat gains and losses [98]. Heat transfer to a zone is pational program are known in the design stage. Then based on those
commonly modeled by using heat conduction equation, heat balance assumptions a single zone model was developed to estimate, in dynamic
method, weighting factor (response factor or transfer function) method, simulation, using thermal balance for normalized outdoor conditions.
and thermal-network method [21] and [99]. This effectively rejected the disturbance caused by the weather
6
conditions and enabled tracking of the optimal set-point. However, the Numerous pieces of literature are available on heating/cooling coil
model did not consider the moisture transmission which is an important modeling with the majority of them are based on governing equations.
element in delivering thermal comfort. All models of heating/cooling coils can basically be categorized into
The majority of previous research works used the Single-Input- two: finite difference and lumped parameter models. The finite differ-
Single-Output (SISO) type model for easier manipulation using linear ence approach results in a comparatively large number of equations
controller [102]. In many cases, building thermal effect is partially or appropriate only for numerical simulation [78]. Most of the previous
fully neglected by the zone/indoor space model. For instance, the models on heating/cooling coils used finite difference approach. For
model developed by Wang et al. [103–106] for indoor space based on instance, Myers et al. [110] and Kabelac [111] used the governing
the conservation of energy and applied thermal balance equation did differential equations to evaluate the dynamic response of a coil as-
not consider the effect of space heat gain or heat loss through the walls, suming that one fluid has an infinite capacitance. Also, Zaheer-Uddin
windows, doors and roof, by air leakage or by operations within the and Zheng [112] developed a transient model for cooling and dehu-
space as well as heat gain from the occupancy, lighting and indoor midifying coils to determine the heat and mass transfer processes oc-
equipment. Platt et al. [97] developed an HVAC zone model assuming curring in the coils although no experimental validation was done by
that the supply air is evenly distributed throughout the whole zone, and them. On the other hand, Karkamaz [113] proposed a numerical model
adjacent zones have the same temperature. However, the thermal sto- based on quasi-steady-state heat and mass transfer analysis to de-
rage of all walls was ignored in this study. Their modeling and pre- termine the short-term energy consumptions and energy performance
diction results demonstrated that the shorter the prediction time, the of the heating and cooling coils for various inlet conditions. Hill and
more accurate the prediction of indoor temperature that can be Jeter [114] suggested a linear sub-grid cooling and dehumidification
achieved. Another study pointed out that highly non-linear heat and coil model focusing on mass transfer. Also, Bocanegra and Khan [115]
moisture transfer are significantly simplified with a dynamic model in and Khan [116] developed models to analyze the performance of a
buildings [107]. On the other hand, the model does not consider the counter flow cooling and dehumidification coil. In their numerical so-
‘actual’ variation of thermal and moisture storage in buildings’ mate- lution, the complex counter crossflow configuration in an actual coil
rials and there are often restrictions in applying the models to short- or was resolved by assuming pure counterflow geometry. Recently, Yao
long-termed behaviors. Goyal et al. [108] constructed a model with et al. [76] built a dynamic model to analyze the thermal qualities of
Resistance Capacitance (RC) networks representing the convective in- cooling coils and later solved using the method of classical control
teraction among a pair of zones. Heat exchange among zones due to theory. This study focused on the effect of perturbations of relevant
radiation was neglected during model development. It was established parameters on the thermal quality of cooling coils under different initial
by the study that the zone temperature prediction capability of RC conditions. A major drawback of this model was that their applied
model considering heat transferred by convection along with conduc- methodology involved comprehensive information on the structure of
tion is closer to the measured temperature compared with the con- cooling coil unit (CCU) and physical properties of fluids, such as fin and
duction-only model. However, applying simplified RC model to con- tube thickness, diameter and spacing, which may not be available from
vective heat transfer among the zones can create model incompatibility manufacturers’ catalog. Therefore, it is quite difficult to implement this
since the uncertain coupling effects between zones can hardly be model in real practice.
identified by the simplified RC models [87]. The lumped parameter approach results in comparatively fewer
equations, however, because of the complex characteristics of heat
5.1.3. The heating and cooling coil model exchangers transient conditions associated with the moving boundary
Heating and cooling coils are another essential element of HVAC between the wet surface region and dry surface region are often ne-
systems. In the AHU, heating and cooling coil act as heat exchangers glected [78]. Wang and Hihara [117] developed a different method
where air loses or gains heat from water passing through the coil. (known as the equivalent dry-bulb temperature method) to predict air
Historically, the primary requirement in coil design was to meet desired coil performance under partially wet and totally wet cooling conditions.
loads with its steady-state performance. However, latterly it has be- However, in their studies due to the difficulties arising during the ex-
come evident that the coil should perform well under dynamic condi- periment to differentiate partially wet cooling with net vapor con-
tions [109]. densate from totally wet cooling, both of the conditions were assumed
The dynamic temperature model of the heating/cooling coil can be to be totally wet cooling. This method was well supported by the study
derived from the energy balance in the air–water heat exchanger of Wang et al. [80]. This study reported that the complex heat and mass
[54,72]. The energy and mass balance equations for the heating and transfer characteristics of CCU could be further simplified without
cooling coil can be expressed as: compromising the prediction accuracy.
Energy balance equations Jin et al. [77] developed a dynamic CCU model by extending the
dTco cooling coil unit engineering model and combining the model with the
Cah = fsw ρw Cpw (Twi − Two ) + (UA)a (To − Tco) + fsa ρa Cpa (Tm − Tco) mass-energy balance equations. It was concluded that this simple
dt
modeling method exactly captured the non-linear characteristics over a
(11)
wide range of system operation. However, in their experiment, they
Mass balance equations only used fresh air instead of mixed air to keep the inlet air temperature
dWco constant. In real practice, it contradicts with the concept of efficient
Vah = fsa (Wm − Wco) running of air handling unit. Moreover, in order to simplify the calcu-
dt (12)
lation, their model was developed based on linear approximation of the
During modeling of heating/cooling coil, Okochi and Yao [22] as-
dry and wet cooling coil unit.
sumed that:
A simple steady-state heating and cooling coil model was developed
by Nassif et al. [73] based on the toolkit of Brandemuehl et al. [118].
• The mass flow rates of both water and air remain constant. They combined the self-tuning mode with the steady-state physical
• The thermal resistance of the coil is negligible. model which was found to better perform in terms of prediction accu-
• The thermodynamic properties of the fluids are independent of the racy than pure physics based model. However, the latent heat transfer
temperature variations for the operating temperature range.
of the cooling coil was not considered in this model. Also, the dynamic
• The coil mass temperature remains constant within the coil, and behaviors of the cooling and heating coils were characterized by con-
• Steady state heat transfer between the water and air. sidering only the steady-state supply-air temperature and using a single
time constant.
7
ṁ a Cpw
5.1.4. The humidifier model When < the NTU value according to Merkel theory can be
ṁ w ⎛ dhs ⎞
Humidification is a mass transfer process of water vapor to atmo- ⎝ dTw ⎠
spheric air, which results in an increase of water vapor in the mixture. given by
Humidification is a necessity in some areas due to the very low hu-
ṁ w
midity that can be present in even a cooling mode in winter [91]. The NTU = MeM
measurement and regulation of moisture in the air is an important ṁ a (21)
phase of air conditioning. The energy and mass balance equations for
Several literatures have been found on cooling tower modeling with
the humidifier model developed by Kasahara et al. [119] can be ex-
most of the earlier research studies focusing on the steady-state op-
pressed as
eration of cooling towers. Merkel's cooling tower model have been well
Energy balance equation for the humidifier model
recognized as the basis for most research studies on cooling-tower
dTh modeling and analysis later (Ninić and Vehauc [120]; Maiya [121]). In
Ch = fsa Cpa (Tsi − Th ) + αh (To − Th )
dt (13) Merkel's model both sensible and latent heat transfer processes were
Mass balance equation for the humidifier model taken into consideration, however, water loss due to evaporation and
drift and heat transfer through the tower walls were neglected, and the
dWh h (t ) Lewis number was assumed as unity to simplify the analysis [122].
Vh = fsa (Wsi − Wh ) +
dt ρa (14) Different studies have assumed different values of Lewis number
based upon different conditions. Poppe and Rögener [123] quoted the
definition of the Lewis relation according to Bošnjaković [124], i.e.
5.1.5. Cooling tower model
A cooling tower working as a primary system component of HVAC (ωs, w +d )/(ωa +d ) − 1 ⎤
system performs the function of heat rejection from the chiller con- Lef = Le 2/3 ⎡
⎢ ln(ωs, w +d )/(ωa +d ) ⎥ (22)
denser [99]. Heat rejection is achieved via heat and mass transfer oc- ⎣ ⎦
curring from direct contact between hot-water droplets from the chiller
where Le is the Lewis number, assumed as a constant of 0.865 and d is
condenser water and relatively cool, ambient air [91]. A cooling tower
the ratio of the molecular weight of water to the molecular weight of
generally comprises fans, a hot-water distribution system, spray noz-
air, which is a constant of 0.622. Grange [125] and [126] claimed that
zles, fill (packing), a collection basin, and a condenser pump [61].
for a wet cooling tower, Eq. (35) is approximately 0.92. It was also
Fig. 3 illustrates the schematic of a cross-flow cooling tower. Chiller's
recommended that the equation by Bošnjaković [124] should be used,
energy consumption is directly influenced by the performance of
and a numerical value of 0.92 be preferred when the fill performance
cooling tower. Moreover, tower fans consume electricity directly [99].
test data is insufficient to predict the Lewis relation of a particular fill
The static diagram (Fig. 4) presented by Li et al. [34] shows the changes
accurately. Kloppers and Kröger [127] specified that, if the ambient air
of total power (consumed by the chiller compressor and tower fan) with
is very humid, the variation of the Lewis relation has little influence on
the variation of cooling tower fan speed with optimal fan speed and
the water outlet temperature; while for dry conditions, a variation of
power consumption 250.351 Hz and 231,174 W, respectively.
the Lewis relation can lead to significantly different results.
In 1925, Merkel developed the first practical theory for cooling
Stabat and Marchio [128] presented a simplified model for indirect
tower modeling, including the differential equations of heat and mass
cooling towers behavior on the basis of Merkel's theory considering the
transfer. His cooling tower model was based on several critical as-
variable water-film temperature along the coil. It was concluded by this
sumptions:
study that the model is capable of predicting energy and water con-
sumptions by the cooling tower under different operating conditions
1. The Lewis factor, Lef, relating heat and mass transfer is equal to 1;
such as variable wet-bulb temperatures or variable airflow rates.
2. The air exiting the tower is saturated with water vapor, and it is
Li et al. [129] developed a Modelica-based dynamic model for a
characterized only by its enthalpy;
typical mechanical draft counterflow cooling tower by simulating its
3. The reduction of water flow rate by evaporation is neglected in the
transient behavior for three different inlet conditions of tower: fixed
energy balance
condition, change in evaporator inlet water temperature, and change in
ambient air condition. Afterward its performance was assessed with the
The effectiveness-NTU method is based on the same simplifying
experimental data from the existing literature. The effectiveness of an
assumptions as the Merkel's model. According to the effectiveness-NTU
extremum seeking control (ESC) scheme for energy efficient operation
method the cooling tower effectiveness can be given by:
of chilled-water system was demonstrated by this dynamic simulation
1−e−NTU(1 −ω) model [34].
ε=
1−ωe−NTU(1 −ω) (15)
Warm, Moist Air Out
where
Fan
Ċ
ω = min Distribution
̇
Cmax (16) Hot Water In System
̇ = min(Cpw ṁ w , Cpa, e ṁ a)
Cmin Spray
(17)
Nozzles
̇
Cmax = max(Cpw ṁ w , Cpa, e ṁ a) (18) Fill
Cpa, e UA
NTU = · Ambient Air In Ambient Air In
Cpa Cmiṅ (19) Cold Water Out
Makeup Water In
The heat rejection rate of cooling tower can be expressed by: Collection Basin
Condenser Water
Q̇rej = ε ·Cmin
̇ ·(TCWS − Twb, i ) (20) Pump
Fig. 3. Schematic of a cross-flow cooling tower [61].
8
1
c= ,
K (29)
∆y (∞)
K=
∆u (∞) (30)
A first order model with dead time is suitable for describing a linear
process and is sufficient for PID controller tuning of HVAC system
[136,137]. On the other hand, second order model with dead time is
more appropriate in case of using advanced controllers within the
system [138].
These first and second order models are applicable for SISO systems
as well MIMO systems. Application of these types of models includes
Fig. 4. Static diagram for total power versus cooling tower fan speed (Reprinted from capturing dynamics of versatile processes of HVAC system [132], con-
[34] with permission from Elsevier). trol of VAV air handling units [133], temperature control [134], re-
presenting time-delay and process gain uncertainties (dynamic varia-
5.1.6. Mixing box model tions) and constraints within the system [34,135].
The mixing of air streams usually under steady and adiabatic con- First/second order over-damped process model with dead time have
ditions is a common practice in air-conditioning systems. As a system comparatively simple structure and very few parameters to be de-
optimization strategy, a portion of return air and fresh air coming from termined from the measured data [20]. Due to the abundance of the
outside is mixed in a mixing box and eventually, supplied to the zone literature on first and second-order systems, the controller design pro-
after passing through heating/cooling coils [130]. cess is also established [20,138]. However, the data gathering process is
The energy and mass balance equations used by Tashtoush et al. quite intrusive requiring the pause of the normal operation of the
[54] for mixing box can be expressed as system [20]. These models are only applied for linear and time-in-
Energy balance equation for mixing box variant (LTI) systems. Thus complicated and time-varying system dy-
namics cannot be represented with such models [20].
mr Cpa Tr + mo Cpa To = m m Cpa Tm (23)
Mass balance equation for mixing box 5.2.2. Data mining algorithms
Data mining algorithms such as Artificial Neutral Network (ANN)
mr + mo = m m (24) and Support Vector Machine (SVM) are capable of dealing complex and
The above equation can also be applied to calculate CO2 con- non-linear system dynamics [22].
centration [74] and the humidity ratio [131] of mixed air by simply ANN has been widely used in HVAC system dynamics modeling,
replacing the temperature variables with their corresponding CO2 and prediction, control and optimization [8,44,84,87,88,139–169]. Several
humidity ratio variables respectively. Current applications of ANNs in building HVAC system include pre-
Table 1 gives a summary of main surveyed studies on physics-based dicting the heating/cooling load [139–143,157], the performance of
model focusing on the field of application, method, main task and chillers [165], the performance of cooling towers [164], the building
outcome and accordingly remarks/recommendations have been made. indoor environmental conditions e.g. indoor temperature
[149–153,155,156,158,163], relative humidity [149,153,158] and the
building heating and cooling energy consumption [84,88,144–148].
5.2. Data-driven model
Besides, ANN has been extensively used in controlling and optimizing
the indoor environment e.g. temperature [8,87,159,162,168], relative
There are nine different types of data-driven models as mentioned in
humidity [8,168], CO2 concentration [8], daylight illuminance [159],
Section 4. These models are discussed in this section.
the performance of HVAC components e.g. AHU [161], chiller [160]
and the building energy consumption [8,44,87,159,166–168]. In ad-
5.2.1. Frequency domain model (first and second-order over-damped dition, a significant reduction in energy consumption can be achieved
process with dead time) by applying ANN model in building HVAC systems [168,169].
There are some processes (e.g., dynamics of zone temperature and Recently, Zhao and Magoulès [170] presented a comprehensive
humidity) in the HVAC system that are known as slow moving due to review of recently developed models for predicting the energy con-
the substantial thermal inertia of the system. Such processes can be sumption of HVAC systems and stated that neural network models
modeled using the first and second order (over-damped) models with could have higher running speed and accuracy in contrast with detailed
dead time [132–135] e.g. engineering models. According to Kumar, et al. [171] and Aydinalp-
First order models with dead time Koksal and Ugursal [84], ANNs can be used to predict energy con-
sumption more consistently than traditional simulation models and
Y (s ) K −Ls
G (s ) = = e regression techniques. Several studies have established ANN models’
U (s ) τs+1 (25)
superiority over linear models [153,158] and physical models
Second-order models with dead time [156,163], [87,172] in modeling the nonlinear HVAC systems. How-
ever, in order to build the models using ANN, a significant amount of
Y (s ) 1 controlling parameters are required and above all, suffers from gen-
G (s ) = = 2 e−Ls
U (s ) as + bs + c (26) eralization capability [173].
Using ANN the modeling structures developed by most of the studies
Where
e.g. [140–142,144,145,149,157,158,163] are single-input-single-
(e−jωc L) output (SISO) or multi-input-single-output (MISO) type. Very few stu-
a = [c − real ( )]/ωc2 ,
G (jωc ) (27) dies e.g. [149,153,158] used neural networks to predict both indoor
temperature and relative humidity. Even though in very few studies
e−jωc L indoor relative humidity has been considered as a factor of thermal
( G (jω ) )
c
b = imag , comfort, in most of the studies this factor has not been considered to
ωc (28) avoid the complications that arise from its inclusion. According to
9
Z. Afroz et al.
Table 1
Summary of main surveyed studies on physics-based model.
Component of Reference Field of application Method Main Task Outcome Remarks

HVAC System
Chiller Browne and Vapor-compression Thermal capacitance Predict the dynamic Prediction result was found to be within ± 10% Advanced control algorithms and complex
Bansal [95] liquid chiller approach performance of a chiller component models that reflect refrigerant
migration can be developed to better predict
the performance of chillers as well as to
reduce the need for empiricism in the
simulation.
Wang, et al. [60] Single-stage and two- Mechanistic/ Test the control performance Identified the effect of chiller dynamics on the HVAC system The performance of the model can be
stage centrifugal deterministic of an HVAC system improved by using least numbers of
chillers assumptions.
Lei and Water chiller Mass and energy Study the transient response Found the minimal feasible valve position at a given operation To increase the reliability of this model
Zaheeruddin [69] balance of the refrigeration system frequency for water chiller minimum numbers of assumptions can be
involved.
Zhang, et al. [70] Air-cooled chiller with Partial differentiation Assist the controller design The model reasonably predicted the transients of the chiller
an economized screw equation (PDE) for controlling chiller
10
compressor performance
Zone Ghiaus and Single zone Heat thermal balance Estimate the heating load in Assessed the heat load needed to reject the disturbance The model can be said incomplete because of
Hazyuk [101] dynamic simulation not considering moisture transmission that
has impact on its performance
Platt, et al. [97] Single zone Physical principles Accurately predict the indoor The model was found quite suitable for near-term predictions of The model can be converted to MIMO type to
and circuit theory environmental conditions zone status. better handle the dynamic thermal processes
within the zones
Goyal, et al. Multi-zone RC networks Predict the zone The model predicted the zone temperatures more accurately than a
[108] temperatures more conduction only model
accurately
Heating and/or Karkamaz [113] Heating and cooling Finite difference Determine the performance The model could analyze the short term performance of heating and To be applicable for analyzing the long-term
Cooling Coil coil approach of the heating and cooling cooling coils. performance of these coils, the model could
coils be simplified.
Yao, et al. [76] Cooling coils Classical control Examine the thermal Analyzed the dynamic heat exchange of coils when some
theory qualities of cooling coils perturbations occurred
Wang and Hihara Cooling coil Equivalent dry-bulb Predict the air coil The prediction for the cooling mode over the coil surfaces was
[117] temperature method performance reasonably accurate.
Nassif, et al. [73] Heating and cooling Improve the performance of The steady state heating and cooling coil models with self-tuning The results could be further improved by
coil the prediction model for parameters showed better performance. using dynamic model substituting the existing
heating and cooling coils steady state model.
Renewable and Sustainable Energy Reviews xxx (xxxx) xxx–xxx
Zhang et al. [174] humidity is an important factor contributing to one's building HVAC system [173,180–182] and fault detection and diagnosis
thermal sensation and comfort. Moreover, it affects one's perception of of HVAC systems [183].
the air quality. Ruano et al. [163] added a radial basis function neural Data mining algorithms possess certain features such as easy im-
networks to build an adaptive model to predict the indoor temperature plementation, capability of dealing with noisy data with many variables
of a school building as a single zone. Ferreira et al. [175] verified an and complex nonlinear relations between inputs and outputs [20,22]
ANN based model predictive control at a campus building, and im- that made them user-friendly to many researchers to use it widely in
plemented a discrete branch and bound approach to optimize the en- HVAC control system. However, these models have certain drawbacks
ergy usage. Also, Spindler and Norford [156] built a multi-zone, multi- such as: for both types of data mining algorithms, no physical inter-
node ANN model to predict the indoor temperature in a multi-zone pretation of the developed model is possible [20,22] and the perfor-
residential building where the accuracy of the predictive model was mance degrades when conditions deviate from training and testing
smaller than the ones obtained in other similar studies. conditions [20]. Moreover, these algorithms are suitable for offline
Through extensive review it is noticed that during model model development whereas online implementation of these algorithms
development most of these studies e.g. [139–142,144,145,149,153, is often cumbersome [20].
157,158,163] considered the whole space as a single zone or used a
single room even though very few studies such as [87,154,156] used 5.2.3. Fuzzy logic model
multi-zone. The effects of thermal interactions, such as convective heat The development of fuzzy logic (FL) models involves the im-
transfer between zones, have rarely been addressed by the studies using plementation of the if-then-else statements whose rules are expressed in
single zone. Garnier et al. [154] constructed an ANN multi-zone model- the form of a table or database [184]. These types of models can be
based predictive control strategy to satisfy the thermal comfort index of subdivided into fuzzy adaptive network (FAN), Takagi–Sugeno (T–S),
a non-residential building. Their result showed that the predictive and adaptive-network-based fuzzy inference system (ANFIS). Several
controller considering the heat transfer between the adjacent rooms current applications of fuzzy logic models in HVAC system include
offers improvement in both energy efficiency and thermal comfort. predicting fan motor speed using ANFIS [185], controlling tempera-
Motivated by this, Huang et al. [87] proposed a new ANN model-based ture/indoor conditions using FAN [186,187] and T-S [188].
modeling approach for multi-zone buildings considering the factors According to Angelov and Filev [189], the adaptive nature of T-S
such as mechanical cooling, ventilation, weather conditions and heat model in addition to the highly transparent and compact form of fuzzy
transfer between the adjacent zones. Further, they examined the im- rules makes them a promising candidate for online modeling and con-
portance of considering convective heat transfer among zones, through trol of complex processes competitive to neural networks. This type of
comparing predictive accuracies of single-zone models to a multi-zone model was used in nonlinear adaptive control, fault detection and di-
model and found more accurate prediction results for multi-zone agnostics, performance analysis of dynamic systems, time-series and
model. forecasting, knowledge extraction, intelligent agents, occupants’ beha-
Compared with ANN, the SVM is resistant to over-fitting the data vior and modeling [189].
and has better performance than simple ANN [20,22,173]. Moreover, Models developed with FL are generally very simple and easy to
SVM can find the global optimum solution in the data and provides the understand [20,22]. However, as FL incorporates the operator's ex-
best fit for the data [20,22,173]. On the contrary, SVM suffers from the perience in model design, operators must have comprehensive knowl-
selection of SVM free parameters [176]. An SVM-based approach uses edge of plant and its various operating states. Moreover, FL model
the nonlinearly separable data into higher dimensional feature space development requires a large amount of performance data for training
through a mapping function where it can be separated linearly [20]. T–S, FAN, and ANFIS [22]. The required knowledge and data for FL
According to Lixing, et al. [173] SVM has been developed to solve model development are not often available for many HVAC compo-
pattern recognition and classification problems, with the introduction nents, and hence it is difficult to model these components using FL
of Ɛ-insensitive loss function. Also, the function of SVM can be extended [20,22].
to solve nonlinear regression estimation problems, such as new tech-
niques known as support vector regression (SVR), which have been
shown to exhibit excellent performance. SVM is a supervised learning 5.2.4. Statistical model
method based on kernel functions used for classification and regression. Statistical models are developed on the basis of how a sample of
However, the previous practical results demonstrate that due to the lack data can be generated from a massive dataset by following a particular
of knowledge of the selection of the three parameters (σ, C, and Ɛ) in an trend [22]. Examples of these types of models include single and mul-
SVR model, the model can suffer from poor forecasting accuracy [177]. tivariate regression, output error (OE), Box-Jenkins (BJ), autoregressive
They have constructed the SVM model to estimate the cooling load of integrated moving average (ARIMA), autoregressive exogenous (ARX),
the building HVAC system by measuring data on the cooling load over a autoregressive moving average exogenous (ARMAX) and finite impulse
period to predict the future cooling load. Later, they have determined response (FIR) [13]. The simple input/output relationship of statistical
the global optima of SVM penalty parameter, intensive loss function models was used by Yiu [190] which is as follows:
and kernel function using the ant colony optimization (ACO). Dong β (q−1) γ (q−1)
et al. [178] investigated the feasibility and capability of SVM in α (q−1) y (t ) = −1
u (t ) + w (t )
θ (q ) δ (q−1) (31)
building load forecasting area. According to their research findings,
Neural Network (NN) can never reach a global solution while the so- Depending on the choice of numerator and denominator poly-
lution of SVM dealing with linearly constrained quadratic programming nomials of input u(t), output y(t) and noise w(t) different models such as
is unique and optimal. Further, Xuemei et al. [176] proposed a forecast ARX, ARMAX, ARIMA, FIR, BJ, and OE are formulated [20].
model of cooling load system based on Stimulated Annealing Particle While some of these models such as FIR, OE, and BJ do not consider
Swarm Optimization (SA-PSO) algorithm that combined the advantages the output, the others such as AR and ARMA do not consider the input
of PSO algorithm and SA algorithm. Their numerical simulation results in the input/output equation. Some models e.g., ARX and ARMAX
established the performance superiority of SA-PSO based SVM model consider both the input and output in their structure [20]. Therefore,
over traditional SVM load forecasting model. the ARX and ARMAX models are useful for the design of the closed-loop
Several Current applications of SVMs in building heating, cooling control system. The model ARMAX shows superior control performance
and ventilation system include prediction of building/HVAC system over ARX as this type of model has the capability of integrating the time
energy consumption [178,179], prediction of cooling load of the series of error in the model structure [20].
11
Several researchers [42,191] and [192] have applied black-box days [195], 30 days [158], 36 days [36]), their model was developed
linear parametric structures such as ARX, ARMAX, BJ and OE models to using data collected over long periods (nine months) and predictions
predict room temperature and relative humidity of different office were performed for different step ahead. They used linear parametric
buildings. Mustafaraj et al. [42] investigated different black-box linear models to obtain a simple and low number of model parameters.
parametric structures such as BJ, ARX, ARMAX structure, and OE to However, this caused compromise with the accuracy, especially in the
identify the humidity and thermal behavior of an office in a modern representation of the heat storage or flywheel effect on the in-
commercial building. Their results demonstrated that all models pro- stantaneous load. The study concludes that there is no specific model
vide reasonably good predictions, but the BJ model outperforms the that can characterize indoor temperature and humidity for all four
ARMAX and ARX models. Romero et al. [89] evaluated the effective- seasons perfectly. This statement contradicts with Patil et al. [196]’s
ness of using linear black-box models to predict the chilled-water research finding. According to their finding, the linear regression
temperature in a variable-speed vapor compression chiller system. The models such as ARX, ARMAX, and NNARX are suitable to describe the
aim of their study was to distinguish a simple yet accurate linear black- greenhouse behavior during most of the year. On the other hand,
box model among the model structures of ARX, ARMAX, OE, and BJ. Frausto et al. [192] and Patil et al. [196] developed models using one
Similar to Mustafaraj et al. [42] this study also agreed that the BJ model year of data. Their developed ARX and ARMAX models had higher
best fit for the prediction. On the other hand, Lowry and Lee [193] used orders (i.e. na = 4 and nb = 5). Yiu and Wang [37] developed a generic
different auto regressive models e.g. ARX, OE, and BJ for predicting SISO and MIMO model for AHU using the ARX and ARMAX structures.
indoor air temperature and found that the OE model provides the best Their results established the effectiveness of the MIMO ARMAX model
fit between simulated and measured internal air temperature data. over SISO ARMAX model.
Several existing research studies established that nonlinear models Through critically reviewing some of the existing literature of sta-
(e.g. NNARX, FFBP, and RBF) generally perform better than linear tistical modeling on HVAC system and building temperature control,
models (e.g. ARX and ARMAX) in predicting buildings’ room tem- some particular features of this type of models come out. First, any
perature [151,155,194–198]. However, according to Jiménez et al. temporal, physical or architectural changes of a building have an im-
[199] linear parametric models can be used to estimate the thermal pact upon space temperature, and this ultimately affects input and
parameters of physical models although in other research Jimenez and output variables of the model [42,191,205]. Second, optimum structure
Madsen [200] established that a nonlinear model is required for an and order of the model are often determined through extensive nu-
adequate description of the thermal characteristics. Mustafaraj et al. merical simulations [37,42,205]. Third, there is a risk of increasing the
[153] in their research study made a comparison between a linear prediction error of the model rapidly in case the prediction prolongs
parametric ARX model and a neural network-based nonlinear auto- over a lengthier time interval [42,191,193]. Fourth, despite very few
regressive model with external inputs (NNARX) to predict indoor office researchers built their models using long-period data collected from
temperature and relative humidity. Their results demonstrated that real buildings [42,205,206], most statistical models are trained using
nonlinear NNARX model outperforms the linear ARX model based upon either simulation data or limited measurements from a small quantity of
prediction accuracy. This is because room temperature and relative HVAC components [37,191,193].
humidity are governed by nonlinear diffusion equations [201]. How-
ever, their studies did not include CO2 concentration as a source of 5.2.5. State Space model
internal heat gain. Further in their research, they used conventional PID State Space model is used for system identification. Prívara et al.
controllers which are only reliable if the system parameters do not [207,208] proposed a subspace model for system identification and
change too much. In reality, since the operating conditions of an HVAC building operation performance forecasting. Jiménez and Madsen
system frequently change due to many factors, such as the weather [209] in their research study outlined and compared a variety of linear
variation, change in heating/ cooling load and fluctuation of occupancy and stationary stochastic models used for modeling the thermal char-
level from day to night and from season to season, the system appears acteristics of buildings and building components. It was found by their
highly nonlinear and time variable. Therefore, in order to obtain good research study that while applying non-linear regression model, im-
control performance, the conventional PID parameters should be tuned pulse response function and state space form model for modeling
or should be replaced by upgraded controllers. building physical systems, the continuous–discrete time stochastic state
Moroşan et al. [202] proposed a distributed ARX model predictive space model can provide a strong context in terms of system identifi-
control strategy for a multi-zone building to regulate its temperature cation.
with intermittently operating mode. They found that the distributed This type of model requires a short time to tune and can be effort-
Model Predictive Control (MPC) which considers the thermal interac- lessly extended to MIMO systems [20]. This type of model can be used
tion among zones outperforms the MPC which does not consider to investigate the dynamic characteristics of the system over a wide
thermal interaction. However, statistical models being linear, time-in- range of conditions that is crucial for the control system design [210].
variant can easily lose accuracies when strong nonlinearity and un- In addition, it reflects the relationship between the input and the output
certainties are in existence within the systems [87]. variables in the model in a very clear way and above all, the model can
Statistical models possess certain advantages such as requiring short be expressed in the form of a matrix that is convenient for computer
calculation time, working well under time varying process noise, scope calculation [210]. However, this subspace state-space method, along-
of the model development with limited data. These models also have side with its advantages has some drawbacks such as: the input of the
some major shortcomings such as low accuracy, inability to capture the system might not excite the system on satisfactory number of fre-
nonlinear dynamics of the system which limits the application of these quencies [211], some key feature or characteristics of the physical es-
types of model in HVAC system limited [20,22]. Moreover, MIMO sence of the system might be lost in data, and natural character of the
model identification using these methods requires a large number of data might create substantial statistical problem [153].
parameters to be determined [20]. Very few literatures can be found on state space model implemented
Developing and validating models using a limited range of data are in HVAC system. Kulkarni and Hong [212] proposed state-space
not reliable for predicting room temperature and relative humidity with method to model the building system and thus described the dynamic
high accuracy. The indoor temperature and relative humidity models of characteristics by developing the building simulation by means of im-
an office building were developed by Mustafaraj et al. [42] using dif- plementing the simulation code on MATLABTM. They considered a
ferent methods of the black-box model such as BJ, ARX, ARMAX building as a thermal network. However, their model was rather an
structure, and OE models. Unlike other models which considered incomplete model covering only winter season and hence, only the
shorter period of data (6 days [203], 14 days [204], 21 days [193], 26 heating system was taken into account. Also, Prívara et al. [213] used
12
subspace state-space method to identify the building model for MIMO. associated behavioral models which use occupants’ presence as an
However, their identification process lacked the persistence of the ex- input, have direct influences on building energy consumption. How-
citation or the excitation itself. On the other hand, Yao et al. [210] ever, their model did not take into account the unnatural peaks arising
constructed a state-space model for reviewing the transient behavior of from repeating the same pattern for each occupant and the displace-
the vapor compression liquid chiller. It was concluded by this study that ment of occupants from one zone to another. Besides, Oldewurtel et al.
this model can be applied for fast predictions of dynamic characteristics [6] developed a stochastic model for building climate control and later
of the key parameters, finding control rules and designing good control analyzed it while taking into consideration the uncertainty arising from
system for the refrigeration system, and for developing a dynamic re- utilizing the weather prediction data. While their simulation result has
sponse toolbox for chillers. established the model as a very effective way of controlling building
climate, its performance in real practice still needs to be investigated.
5.2.6. Geometric model Sun et al. [217] developed a stochastic model for overtime occupancy
Geometric models use mathematical methods to model real objects based on measured occupancy data from an office building which
using computer graphics and computer-aided design (CAD) [22]. They shows that this model can be used to generate occupant schedules
represent a system through the use of two-dimensional (2D) or three- during overtime as an input to building energy simulations. Still, there
dimensional (3D) geometric shapes such as curves, surfaces, and vo- is a necessity of improving the model by including the overtime char-
lumes [20,22]. The spline surfaces i.e. Hermite-splines, Bezier-splines, acteristics of occupants on weekends and holidays.
B-splines and the thin plate spline (TPS) are very common tools used in
geometric modeling [22]. The thin plate splines having the smooth 5.2.9. Instantaneous model
features and a good ability of extrapolation can effectively be used to Just in Time (JIT) is an instantaneous model comprising a statistical
interpolate scattered data. The thin plate spline performs a special and a pattern model to find the patterns in previous data similar to the
function to map the input-output data [214]. TPS has particular merits current data [20]. Iino et al. [219] proposed the “hybrid modeling”
and demerits compared to ANN. One of the positive sides of TPS is that approach with physical and JIT model for building thermal load pre-
it is not affected by initial conditions [20,22]. On the other hand, TPS is diction and optimal energy saving control. The JIT model used by this
sensitive to the noise present in the data on which it interpolates and study to predict the thermal load demand curve based on weather
becomes computationally expensive as the data size increases [214]. prediction in the normal case did not work well when a new event took
Not many literatures can be found on this type of model, especially in place which was not included in the past data. Their result showed that
HVAC system the application of this type of model is rare. in summer season JIT model performed good thermal load estimation
as the weather condition was stable; however, on a unique weather day
5.2.7. Case-based reasoning thermal load estimation by this model was not good.
Case-Based Reasoning (CBR) approach is suitable for a nonlinear JIT model can be applied for modeling the other HVAC components
and ill-defined system [20]. These models find similar past cases to such as cooling coil, and chiller from the measured data. In order to
construct the models [215]. The models mature with time and add data implement JIT technique effectively, a large database is required and
to the case-base online to increase the future prediction performance also the search takes a longer period if the number of parameters is
[20]. large [20].
Topological case-based modeling (TCBM) is a distinguishing CBR Table 2 gives a summary of main surveyed studies on data-driven
technique in which the models are built on a subset of data instead of model focusing on the field of application, main task, outcome, and
the whole data [20]. Nishiguchi et al. [214] used TCBM to compress the accordingly remarks/recommendations have been made.
measured raw data as well as to reduce the measurement noise and
thus, minimized the computational cost of generating and revising the 5.3. Gray-box model
model. Tsutsui and Kamimura [216] built a TCBM to monitor and
analyze chiller system performance deterioration. The gray box models which can be regarded as the combination of
While using this model it is easy to distinguish between interpola- white box and black box models, developed from the idea of over-
tion and extrapolation [151]. However, the case-base should be com- coming the shortcomings of both of these models [21]. The basic
prehensive to cover all the input cases [20]. This type of model is not structure of the model is formed from physics-based methods while the
very popular as the model suffers from the problems related to the model parameters are determined by using the parameter estimation
unseen cases [20,22]. algorithms on the measured data of the system. In fact, there are some
physical processes of the HVAC system that are less clearly defined by
5.2.8. Stochastic model the thermodynamic equations and at the same time certain information
Stochastic models can deal with the random processes of the HVAC or adequate training data are lacking. In this case, gray box or hybrid
systems that act as random variables, and these processes can be model is the most effective evaluated to date. This approach is espe-
modeled using the probability density functions (PDF) [20,22]. Ex- cially beneficial for control applications when the model is expressed in
amples of Stochastic model include simulation of occupants’ time a suitable form such as transfer function or state space [90].
schedules and activity level [217,218]. Ghiaus et al. [90] developed a hybrid model in a linear form to
One of the important features of stochastic model is the versatility of represent a constant air volume AHU without dehumidification that can
its application to many physical processes that can be approximated to be used to improve its control function. To achieve separation of ele-
standard normal and uniform distributions although a large amount of mental transfer functions for avoiding coupling between humidity and
data is a prerequisite for that [20,22]. Moreover, the model cannot temperature, the model was developed based on the assumption that
provide accurate predictions if the probability density functions (PDF) each element changes only one variable: variations in temperature
is not modeled properly [20,22]. difference take place in the cooling coils while changes in humidity
The majority of research studies have utilized stochastic models in ratio difference occur in the humidifier. This indicates that the cooling
their simulation work, however, in real applications, their research coil is under dry operating conditions and in the humidifier air tem-
findings may not coincide with actual performance. Page et al. [218] perature is constant. However, in real practice these assumptions are
represented the influence of occupants on the building by stochastic not achievable [21].
models which emphasized on occupants’ time schedule and their in- Afram and Janabi-Sharifi [220] developed a gray-box model for the
teracting level with their indoor environment. They revealed that their residential HVAC system for representing the effects of on/off con-
proposed model (due to related metabolic heat gains), along with other trollers on the energy consumption of the total system. The recent more
13
advanced controllers (e.g., PID, MPC) and energy conservation strate-
Occupants’ schedule can be integrated into the model for
The model suffers from one of the drawbacks as stated
It is necessary to validate the model by a real system.

gies (e.g., set-point reset, passive/active thermal energy storage) often
There is a necessity for investigating the predictive
movement of occupants from one zone to another.
holidays can be included in the model for further

used in existing HVAC system were not integrated into their model.
Occupants’ overtime schedules on weekends and

The model can be improved by considering the
The model is only applicable in winter season.
Vaghefi et al. [221] motivated by the practical limitations of si-
mulation-based optimizations developed a hybrid forecast model from
physics-based and data-driven model to forecast optimal heating and
improving the HVAC control system
cooling set point values. The combined forecast model was then used in
improvement in simulation result.

an MPC framework to control heating and cooling set points that ulti-
mately reduced the total energy cost and thermal discomfort penalty at
strategy in the real system.
the same time. However, it was not an adaptive model. Also, the var-
iation in occupancy level, heat transfer due to inter-correlation between
zones, and zone surface phenomena were not considered by this model.
The gray-box modeling approach proposed by Braun and
Chaturvedi [35] used a transfer function with parameters and was
Remarks
earlier.
constrained to satisfy a simple physical representation for energy flows
in the building structure. The model was used to predict the transient
cooling and heating requirements for the building. The model and
Achieved more accurate prediction results than a
Identified the model to be used in MPC to control
uncertainty of the weather forecast in its control
occupants’ overtime schedule on building energy

Stochastic MPC was able to directly capture the
Achieved a good indoor temperature regulation
Showed improvement in both building energy
training method were extensively tested for different buildings and

Occupants’ presence have direct influence on
Described the dynamic characteristics of the
The model represented the influence of locations using data generated from a detailed simulation program. It
was established by this study that one to two weeks of data is sufficient
Approximately 50% energy was saved
the indoor temperature of building.
to train a hybrid/gray box model to predict transient cooling or heating

that reduced energy consumption
requirements accurately.
efficiency and thermal comfort
building energy consumption
Table 3 gives a summary of main surveyed studies on gray box

model focusing on the field of application, main task and outcome and
accordingly remarks/recommendations have been made.
single-zone model.
consumption.
6. Discussion and summary

decisions
Outcome
building
The whole structure of HVAC plant is very complex where time-

varying system dynamics, slow-moving processes with time delays and
non-ideal behavior of actuators are prevailing. Thus, substantial dis-
Optimization of the controller to maintain the
thermal comfort while minimizing the energy
Optimize the energy usage while maintaining
Improve the accuracy of the building energy
turbances, constraints, and uncertainties are imposed by running the

Satisfy the thermal comfort index of a non-
total HVAC system dynamics. The necessity of appropriate modeling

Predictive control of the air-conditioned
Survey the importance of considering
comes from the requirement of handling disturbances, constraints, and

Identify the model for MIMO system
Identify the patterns of occupancy
uncertainties existing within the HVAC system dynamics. At the very

beginning of this paper HVAC system components and the character-
istics of its system parameters have been introduced and simultaneously
Building climate control
convective heat transfer
indoor thermal comfort
the importance of accurate modeling to deal with this highly non-linear

residential building
system dynamics has been demonstrated. Next, the function of perfor-

mance comparison metrics has been defined. Afterward, the modeling
consumption.
simulations
techniques used in HVAC system has been categorized, defined and

Main Task
system.
critically reviewed to ascertain the application and feature of individual

modeling techniques. To what extent these modeling techniques are
suitable to implement in real system has been revealed through this
Effect of overtime occupancy on
review. Also, the approach, role, outcome and shortcomings of some of


Building temperature control
Building heating and cooling
the existing studies/developed models where these modeling techni-

Indoor temperature control

Effect of occupants on the
Building thermal comfort
ques have been used, are identified. From the above sections the
Building HVAC system
strengths and weaknesses of these modeling techniques have been re-

Field of application
vealed which will be briefly described in this section to perform a

comparative study among these modeling techniques. Besides, studied
Summary of main surveyed studies on data-driven models.
modeling techniques will be compared based on four performance

system
criteria which are very important for selection of modeling techniques

to apply in HVAC system. Lastly, recommendations will be drawn in
this section which will be beneficial for future research studies.
Ferreira, et al. [175]
Garnier, et al. [154]
Prívara, et al. [213]

Ruano, et al. [163]
Kulkarni and Hong

Huang, et al. [87]
Page, et al. [218]
Oldewurtel, et al.
Sun, et al. [217]
6.1. Comparison of modeling techniques based on weakness and strength

Reference
Based on the extensive literature review the weaknesses and

[212]
strengths of three basic modeling techniques used in HVAC systems are

[6]
identified which is important to know during model selection. From the

above sections it is established that physics-based modeling technique
uses a large number of mathematical equations derived from physical
Types of Data-
driven Models
State Space
and chemical laws of conservation. Therefore, it is necessary to un-

Stochastic
derstand the physical behavior of the parameters and operational pro-

Table 2
ANN
cesses of the system to develop this type of model. Since this type of
model is developed based on certain assumptions its prediction
14
capability is not so accurate and ultimately when the system is con-
Due to lack of combination between temperature and humidity in the cooling

coil and humidifier the model can be said not practical enough to use in real
structed and operated, it does not perform according to its design in-
The model can be upgraded by integrating more advanced controllers and
The performance of the model can be improved by taking into account

tent. In addition, detailed physics-based modeling is very complex and
involves iteration processes to develop. On the contrary, this type of
model holds some positive features such as ease of analysis, robust
occupants’ time schedule and thermal interaction among zones.

generalization capability and requirement of less training data.
From the discussion in Section 6.1, some major shortcomings of
physics-based models associated with research studies are revealed
which are as follows:
energy conservation strategy into the model.
• Some of the models provided very poor prediction accuracy due to

the involvement of excess numbers of assumptions. As a result, the
resulting design did not perform in accordance with the model.
• Some of the models can be expressed as insufficient because of not
taking into account some important parameters or operating con-
ditions with respect to those models. On the other hand, some
models have been developed considering all necessary physical
parameters and operating conditions. As a result, the dynamic re-
lationship between the system and the ambient disturbances were
better identified by these models, and provided much more in-
Remarks
system.
formation about the system dynamic behaviors. However, while

executing, those models were found too complex to be implemented
in real-time, affordable systems.
The obtained set point values reduced both total energy
cost and thermal discomfort penalty at the same time.
• Most of the developed models were SISO type which is unable to

The developed models had high accuracy and were
represent the multivariable interactions in HVAC system.

•
able to predict the output variables precisely.
Developed a constant-air-volume AHU model
Some of the models were found steady-state or quasi-steady-state

which is not suitable for handling high-frequency disturbances and
dealing various HVAC design problems.
When sufficient training data is available and inputs are within the
training range, black box modeling technique can provide high accu-
racy, and also its simple structure makes them applicable for real-time
operation and control. This type of models completely relies on training
data and often suffers from poor generalization capability. To develop
this type of model it is not necessary to have a good understanding of
Outcome
the system physics.

From the discussion in Section 6.2, some major limitations of dif-
ferent types of data-driven models associated with research studies of
Represent the effects of on/off controllers on the
Improve the control of constant-air-volume AHU
HVAC system and building thermal comfort are identified and listed
below:
energy consumption of the HVAC system
• To avoid complicacy within the model some key parameters or

Control heating and cooling set points
operating conditions were not considered in some models.

• Many models have been developed considering the whole building a
single zone to avoid the phenomena of thermal interaction among
adjacent zones.
• It is hard to achieve good prediction accuracy using statistical model
in HVAC system or building indoor air quality control although this
type of model has performed successfully to describe the greenhouse
Summary of main surveyed studies on Gray box/Hybrid model.
Main Task
behavior during most of the year. Furthermore, using a limited

range of data in model development is not reliable to predict e.g.,
indoor temperature, relative humidity, etc.
• In some cases the model has not been validated by applying in a real
Heating/cooling coil
Field of application
system.
Residential HVAC
• The identification process using the subspace state-space method

may lack the persistence of the excitation or the excitation itself.
• The model describing the dynamic characteristics of the building
system
AHU
can be considered incomplete when seasonal variation is not cov-

ered by the data.
Vaghefi, et al. [221]
Afram and Janabi-
Ghiaus, et al. [90]
Gray-box/hybrid model is developed from the necessity of over-

Sharifi [220]
coming the shortcomings of both black-box and white-box model.

Therefore, this type of model holds many benefits such as high accu-
Reference
racy, easy generalization capability, less complexity and low compu-

Table 3
tational cost and ultimately, can deliver good control performance.

However, to develop a gray-box/hybrid model more effort is required
15
as the model development process involves the implementation of both low level is desirable. However, depending on the objective of research
governing equations and a large amount of training data. study and available facility any of the performance criteria can be
From the discussion in Section 6.3 some major shortcomings of compromised and accordingly suitable modeling technique can be
hybrid models associated with research studies of HVAC systems and identified. For example at the design stage of HVAC system training
building thermal comfort are pinpointed and listed below: data is not available. In that case physics-based modeling technique is
the only way to predict and analyze the HVAC system dynamics. On the
– Some of the models were developed based on some assumptions other hand, data-driven model requires high training data which is only
which are not achievable in real system. possible to record when the system is present and running. Still, the
– Some of the models can be considered as incomplete because tar- data-driven modeling technique is very popular because of its simplicity
geted at simplification some of the important factors e.g. variation and generally providing high accuracy in the prediction although this is
in occupancy level, heat transfer among adjacent zones etc. of in- not true for all types of data-driven model. Yet, sometimes it necessi-
door thermal comfort were not taken into account during model tates for improving the performance of the system through balancing
development. between energy consumption and thermal comfort level which is not
– Sometimes the model does not hold the flexibility to use in other possible with a data-driven model when sufficient training data is not
HVAC systems. available. In that case, gray box model is very effective. Table 5 gives a
clear picture about the main features of these modeling techniques.
Table 4 provides a comparative study of three basic modeling High prediction accuracy is given by gray-box and some types of
techniques on the basis of their weaknesses and strengths. data-driven model followed by other types of data-driven model except
the statistical type (Table 5). The prediction capability of physics-based
6.2. Based on performance criteria and statistical model is low. On the other hand, physics-based model
holds high generalization capability, followed by gray-box and some
This section analyzes the performance of three basic modeling types of data-driven model. Other types of data-driven model possess
techniques by making a comparison based on four performance criteria: medium generalization capability. The structure of gray-box is not so
prediction accuracy, generalization capability, training data require- simple or complex and necessitates medium amounts of training data
ment, complexity level that are very important in terms of model se- during model development compared with the other two modeling
lection. During model selection it is desired that prediction accuracy techniques. On the contrary, physics-based model requires low amounts
and generalization capability are high leveled while for other two of training data and its complexity level is high if all necessary physical
performance criteria: training data requirement and complexity level, parameters and operating conditions are considered while data-driven
Table 4
A comparative study of three major modeling techniques based on weakness and strength.
Modeling Technique Weakness Strength Research Studies
Physics based (white box/ • Complex [21,25,222] • Easy to analyze [21,22] Lü [223]; Ghiaus and Hazyuk [101]; Badescu and Sicre
mathematical/ forward)
model
• Iteration
[25,32].
process is often required • Represents
[20]
the system more closely [224]; Badescu and Staicovici [225]; Badescu, et al.
[226]; Scotton, et al. [227]; etc.
• Completely relies on the knowledge of
the process and the physical laws
• Have better generalization
capability [20,86,220]
governing the process [20,25] • Requires less training data [25]
• Requires specification of many physical
parameters [25,35,89]
• Involves assumptions that may not hold
in practice [25,222].
• Requires understanding of the system
physics [86,220]
• Poor accuracy [86,220].
Data driven (black box/ • Less flexible [32] • Simple [25,32] Yiu and Wang [37]; Mustafaraj, et al. [42]; Romero,
empirical/ inverse) model • Require a significant amount of training
data [25,35].
• Coincides with real-time operation
and control [21]
et al. [89]; Chen, et al. [56]; Kusiak, et al. [44]; Hou,
et al. [229]; Hu, et al. [230]; Afram and Janabi-Sharifi
• physical
May not always reflect the actual
behavior [25,227]
• Does not require understanding of
the system physics
[86]; Chintala and Rasmussen [41]; Wei, et al. [8]; etc.
• Can give unrealistic or non-physical [21,25,86,153,220]

results when the inputs are outside the
training range [20,21,25,228]
• Provides good prediction accuracy
[32,86,220]
• Suffer from the poor generalization
capabilities [86,220]
• Most of these models cannot ensure
stable performance prediction [25,35].
• Entirely dependent on the measurement
data of the input and output variables
[20,86,153,220]
Gray box (hybrid) model • Requires more effort to develop [22,86] • Possesses high accuracy [22,220] Afram and Janabi-Sharifi [220]; Braun and Chaturvedi
• Accuracy strongly depends on the • Easy to generalize [22,220] [35]; Leephakpreeda [231]; Zhao, et al. [58]; Vaghefi,
richness of data used to train the model
[25]
• Well adapted to perform
optimization [228]
et al. [221]; McFadden, et al. [232]; Ghiaus, et al. [90];
etc.
• Comparatively less complex [25]
• Comparatively
costs [25]
less computational
• Covers limited extrapolation outside

the range of the training data [25].
• May provide good control
performance [25]
16
Table 5
Comparison of three basic modeling techniques based on performance criteria.
Modeling Technique Prediction Generalization Training Data Complexity Level Research Studies
Accuracy capability Requirement
Physics based (white box/ mathematical/ L H L H [20–22,25,84,86,146,220,233]

forward) model
Data driven (black box/ empirical/ inverse) H/ M/ La L/ Mb H L
model
Gray box (hybrid) model H M M M
Note: The letter H, M, and L stand for High, Medium and Low respectively.
a
Prediction accuracy of data-driven models usually depends on its type e.g. data mining algorithm, fuzzy logic, state-space and stochastic models gives high prediction accuracy;
frequency domain with dead time, geometric, case reasoning and instantaneous models give medium prediction accuracy; only statistical model provides low prediction accuracy.
b
Generalization capability of data-driven models is commonly medium to low depending upon its type e.g. frequency domain with dead time, data mining algorithm, state-space,
stochastic and instantaneous models possess high generalization capability while fuzzy logic, statistical, geometric and case reasoning models possess medium generalization capability.
model is opposite to that. That means data-driven model requires huge overcome by generating artificial data that already contains desired
amounts of training data and its structure is quite simple. properties.
– It is very important to cover the seasonal variation by the data while
6.3. Recommendation developing any model that represents building dynamic conditions.
– Among different types of data-driven modeling technique appro-
From a critical survey of the literature on different models used in priate model should be selected based on their features and ease of
HVAC system, it is revealed that almost every model has a major/minor application. It is also important to choose between linear and non-
shortcoming generated from either assumptions or unmeasured dis- linear depending on the purpose of the model. For example, to es-
turbances or any uncertainties in some properties. It is a very challen- timate the thermal parameters of physical models linear parametric
ging task to develop an accurate and effective model for the HVAC models can be used while for getting an adequate description of the
system without any downside. However, if the limitations of developed thermal characteristics nonlinear model is required.
models used in HVAC system can be identified properly and minimized
accordingly, these models can provide better results and effectively take
part in building energy conservation in the near future. Some points 6.3.3. Hybrid model
have been suggested below to overcome the limitations of these models.
These points may act as guidelines for future research studies in this – The developed model must be free from any assumptions that may
area. lead the model to uncertainty to implement in real system.
The identified shortcomings of physics based, data-driven and hy- – Since the ultimate objective of HVAC system is to provide thermal
brid models can be overcome to some extent by taking into con- comfort, therefore during model development the important factors
sideration some important points during model development which are of thermal comfort must not be overlooked.
listed below: – The model must be adaptive in real system.
– During model development recent energy conservation strategies
6.3.1. Physics-based model and advanced controllers of HVAC system should be integrated.
– While developing physics-based models attention should be given

on using least numbers of assumptions. 7. Conclusion
– The physics-based model should be simplified or detailed based on
the objective of any research studies. However, important para- To improve the energy efficiency of buildings, appropriate appli-
meters or operating conditions that have major impacts on the cation of advanced control strategies in Heating, Ventilation, and Air-
performance of that model should not be overlooked. conditioning (HVAC) systems is important. Selection of suitable mod-
– The models for HVAC system should be MIMO type especially the eling techniques and appropriate modeling of the whole HVAC system
zone model so that it can handle the dynamic thermal processes are two vital tasks on the way to improve the functionality of the
within the zones and the interaction with the environment because control system, lessen building energy consumption and enhance in-
of the continuous variation of the external and internal environment door thermal comfort. Therefore, in-depth knowledge on the role, ap-
of the zone. plication, merits, shortcomings and outcomes of different modeling
– The developed model must be applicable and relate with the concept techniques used in HVAC systems is important for the selection and
of efficient running of HVAC components in real system. design of an appropriate model. From an extensive literature review,
this study identifies the strengths and weaknesses of various modeling
6.3.2. Data-driven model techniques and the shortcomings of some of the existing studies/de-
veloped models, resulting in recommendations. Comparing three basic
– During model development, relevant important parameters and modeling techniques based on the performance criteria, the study found
operating conditions must be taken into account. that each modeling technique has both positive and negative features
– While developing model for multi-zone building, the model should for consideration when selecting a technique for model development.
not be simplified to single zone to ignore the effect of thermal in- For example, in physics-based modeling minimizing the number of as-
teraction among adjacent zones. sumptions results in better outcomes while the data-driven models
– A statistical model is not suitable where high prediction accuracy is benefit from larger datasets.
required. Following the recommendations, this study could be of use as a
– The proposed model should be validated by a real system. guideline for the building research community or energy management
– This study also aligns with the statement drawn by Prívara et al. personnel in order to select the most suitable modeling technique de-
[213] that is the drawback of state space model can be possibly pendent on the objective of the research study or available facility.
17
Acknowledgements Fundamentals vol. SI. Atlanta: ASHRAE, 2013.

[33] R. McDowall, Fundamentals of HVAC Systems, 2007.
[34] Li X, Li Y, Seem JE, Li P. Dynamic modeling and self-optimizing operation of
This research is financially supported by the Murdoch University chilled water systems using extremum seeking control. Energy Build
Research Scholarship (MURS). This study also acknowledges the 2013;58:172–82.
Murdoch University Energy Management Facility for providing neces- [35] Braun JE, Chaturvedi N. An inverse gray-box model for transient building load
prediction. HVACR Res 2002;8:73–99.
sary resources for conducting research at their premises. [36] Ríos-Moreno GJ, Trejo-Perea M, Castañeda-Miranda R, Hernández-Guzmán VM,
Herrera-Ruiz G. Modelling temperature in intelligent buildings by means of au-
References toregressive models. Autom Constr 2007;16:713–22.
[37] Yiu JC-M, Wang S. Multiple ARMAX modeling scheme for forecasting air con-
ditioning system performance. Energy Convers Manag 2007;48:2276–85.
[1] Pérez-Lombard L, Ortiz J, Pout C. A review on buildings energy consumption in- [38] C. Agbi, Z. Song, and B. Krogh, Parameter identifiability for multi-zone building
formation. Energy Build 2008;40:394–8. models. In: Proceedings of the 2012 IEEE 51st IEEE Conference on Decision and
[2] Zhao J, Zhu N, Wu Y. The analysis of energy consumption of a commercial building Control (CDC), 2012, pp. 6951–6956.
in Tianjin, China. Energy Policy 2009;37:2092–7. [39] Li X, Wen J. Review of building energy modeling for control and operation. Renew
[3] Lin H-W, Hong T. On variations of space-heating energy use in office buildings. Sustain Energy Rev 2014;37:517–37.
Appl Energy 2013;111:515–28. [40] Afram A, Janabi-Sharifi F. Theory and applications of HVAC control systems – A
[4] Fasiuddin M, Budaiwi I. HVAC system strategies for energy conservation in com- review of model predictive control (MPC). Build Environ 2// 2014;72:343–55.
mercial buildings in Saudi Arabia. Energy Build 2011;43:3457–66. [41] Chintala RH, Rasmussen BP. Automated multi-zone linear parametric black box
[5] Rahman MM, Rasul MG, Khan MMK. Energy conservation measures in an in- modeling approach for building HVAC systems. In: ASME 2015 Dynamic Systems
stitutional building in sub-tropical climate in Australia. Appl Energy and Control Conference, 2015, pp. V002T29A004-V002T29A004.
2010;87:2994–3004. [42] Mustafaraj G, Chen J, Lowry G. Development of room temperature and relative
[6] Oldewurtel F, Parisio A, Jones CN, Gyalistras D, Gwerder M, Stauch V, et al. Use of humidity linear parametric models for an open office using BMS data. Energy Build
model predictive control and weather forecasts for energy efficient building cli- 2010;42:348–56.
mate control. Energy Build 2012;45:15–27. [43] Rodriguez E, Rasmussen BP. A nonlinear reduced-order modeling method for dy-
[7] Široký J, Oldewurtel F, Cigler J, Prívara S. Experimental analysis of model pre- namic two-phase flow heat exchanger simulations. Sci Technol Built Environ
dictive control for an energy efficient building heating system. Appl Energy 2016;22:164–77.
2011;88:3079–87. [44] Kusiak A, Li M, Tang F. Modeling and optimization of HVAC energy consumption.
[8] Wei X, Kusiak A, Li M, Tang F, Zeng Y. Multi-objective optimization of the HVAC Appl Energy 2010;87:3092–102.
(heating, ventilation, and air conditioning) system performance. Energy [45] Knabe G, Le H. Building simulation by application of a HVAC system considering
2015;83:294–306. the thermal and moisture behaviors of the perimeter walls. IBPSA Build Simul Proc
[9] Zeng Y, Zhang Z, Kusiak A. Predictive modeling and optimization of a multi-zone 2001;2:965–72.
HVAC system with data mining and firefly algorithms. Energy 2015;86:393–402. [46] J. L. M. Hensen, On the thermal interaction of building structure and heating and
[10] Aggelogiannaki E, Sarimveis H. A simulated annealing algorithm for prioritized ventilating system: Technische Universiteitt Eindhoven, 1991.
multiobjective optimization - Implementation in an adaptive model predictive [47] Chow T, Clarke J, Dunn A. Primitive parts: an approach to air-conditioning com-
control configuration. IEEE Trans Syst Man Cybern Part B: Cybern ponent modelling. Energy Build 1997;26:165–73.
2007;37:902–15. [48] Lam JC, Hui SC, Chan AL. Regression analysis of high-rise fully air-conditioned
[11] Zhang X, Fong KF, Yuen SY. A novel artificial bee colony algorithm for HVAC office buildings. Energy Build 1997;26:189–97.
optimization problems. HVAC R Res 2013;19:715–31. [49] J. Cui, T. Watanabe, Y. Ryu, Y. Akashi, and N. Nishiyama, Numerical simulation on
[12] Hu M, Weir JD, Wu T. Decentralized operation strategies for an integrated building simultaneous control process of indoor air temperature and humidity. In:
energy system using a memetic algorithm. Eur J Oper Res 2012;217:185–97. Proceedings of the Sixth international IBPSA conference, proceedings; 1999, pp.
[13] Coelho L Dos Santos, Askarzadeh A. An enhanced bat algorithm approach for re- 1005–1012.
ducing electrical power consumption of air conditioning systems based on differ- [50] J.-P. Bourdouxhe and P. André, Simulation of a centralized cooling plant under
ential operator. Appl Therm Eng 2016;99:834–40. different control strategies. In: Proceedings of the International Conference on
[14] Henze GP, Dodier RH, Krarti M. Development of a predictive optimal controller for Building Performance Simulation (IBPSA097); 1997, pp. 95–102.
thermal energy storage systems. HVAC R Res 1997;3:233–64. [51] Swider DJ. A comparison of empirically based steady-state models for vapor-
[15] Geem ZW, Kim JH, Loganathan GV. A new heuristic optimization algorithm: compression liquid chillers. Appl Therm Eng 2003;23:539–56.
harmony search. Simulation 2001;76:60–8. [52] Keniar K, Ghali K, Ghaddar N. Study of solar regenerated membrane desiccant
[16] Fong KF, Hanby VI, Chow TT. A robust evolutionary algorithm for HVAC en- system to control humidity and decrease energy consumption in office spaces. Appl
gineering optimization. HVAC R Res 2008;14:683–705. Energy 2015;138:121–32.
[17] Kusiak A, Tang F, Xu G. Multi-objective optimization of HVAC system with an [53] Ahn B-L, Yoo S, Kim J, Jeong H, Leigh S-B, Jang C-Y. Thermal management of LED
evolutionary computation algorithm. Energy 2011;36:2440–9. lighting integrated with HVAC systems in office buildings. Energy Build 2016.
[18] Xi X-C, Poo A-N, Chou S-K. Support vector regression model predictive control on a [54] Tashtoush B, Molhim M, Al-Rousan M. Dynamic model of an HVAC system for
HVAC plant. Control Eng Pract 2007;15:897–908. control analysis. Energy 2005;30:1729–45.
[19] Greene AM. Principles of Heating, Ventilating and Air Conditioning. New York: J. [55] Zhang D, Xia X, Cai N. A dynamic simplified model of radiant ceiling cooling
Wiley; 1936. integrated with underfloor ventilation system. Appl Therm Eng 2016.
[20] Afram A, Janabi-Sharifi F. Review of modeling methods for HVAC systems. Appl [56] Chen X, Wang Q, Srebric J. A data-driven state-space model of indoor thermal
Therm Eng 2014;67:507–19. sensation using occupant feedback for low-energy buildings. Energy Build
[21] Homod RZ. Review on the HVAC system modeling types and the shortcomings of 2015;91:187–98.
their application. J Energy 2013;2013. [57] Wang S. Dynamic simulation of a building central chilling system and evaluation
[22] Okochi GS, Yao Y. A review of recent developments and technological advance- of EMCS on-line control strategies. Build Environ 1// 1998;33:1–20.
ments of variable-air-volume (VAV) air-conditioning systems. Renew Sustain [58] Zhao Y, Zhou S, Li L. Dynamic characteristics modeling of a heat exchanger using
Energy Rev 2016;59:784–817. neural network. In: Proceedings of the First International Conference on Intelligent
[23] Harish VSKV, Kumar A. A review on modeling and simulation of building energy Networks and Intelligent Systems, 2008. ICINIS'08; 2008, pp. 13–18.
systems. Renew Sustain Energy Rev 2016;56:1272–92. [59] Parvaresh A. A new mathematical dynamic model for HVAC system components
[24] Dounis AI, Caraiscos C. Advanced control systems engineering for energy and based on Matlab/Simulink. Int J Innov Technol Explor Eng 2012;1:1–6.
comfort management in a building environment–a review. Renew Sustain Energy [60] Wang S-W, Wang J, Burnett J. Mechanistic model of centrifugal chillers for HVAC
Rev 2009;13:1246–61. system dynamics simulation. Build Serv Eng Res Technol 2000;21:73–83.
[25] Wang S, Ma Z. Supervisory and optimal control of building HVAC systems: a re- [61] Li P, Qiao H, Li Y, Seem JE, Winkler J, Li X. Recent advances in dynamic modeling
view. HVAC R Res 2008;14:3–32. of HVAC equipment. Part 1: Equipment modeling. HVACR Res 2014;20:136–49.
[26] Afram A, Janabi-Sharifi F. Theory and applications of HVAC control systems–A [62] Bendapudi S, Braun JE. A review of literature on dynamic models of vapor com-
review of model predictive control (MPC). Build Environ 2014;72:343–55. pression equipment: publisher not identified; 2002.
[27] Shaikh PH, Nor NBM, Nallagownden P, Elamvazuthi I, Ibrahim T. A review on [63] Ghahramani A, Tang C, Becerik-Gerber B. An online learning approach for quan-
optimized control systems for building energy and comfort management of smart tifying personalized thermal comfort via adaptive stochastic modeling. Build
sustainable buildings. Renew Sustain Energy Rev 2014;34:409–29. Environ 2015;92:86–96.
[28] Trčka M, Hensen JL. Overview of HVAC system simulation. Autom Constr [64] Ma Y, Matuško J, Borrelli F. Stochastic model predictive control for building HVAC
2010;19:93–9. systems: complexity and conservatism. IEEE Trans Control Syst Technol
[29] Subbaram Naidu D, Rieger CG. Advanced control strategies for heating, ventila- 2015;23:101–16.
tion, air-conditioning, and refrigeration systems - an overview: Part I: hard control. [65] Kohlenbach P, Ziegler F. A dynamic simulation model for transient absorption
HVAC R Res 2011;17:2–21. chiller performance. Part I: the model. Int J Refrig 2008;31:217–25.
[30] Subbaram Naidu D, Rieger CG. Advanced control strategies for HVAC&R systems - [66] Bendapudi S, Braun JE, Groll EA. Dynamic model of a centrifugal chiller sys-
an overview: Part II: soft and fusion control. HVAC R Res 2011;17:144–58. tem–model development, numerical study, and validation. ASHRAE Trans
[31] Mirinejad H, Welch KC, Spicer L. A review of intelligent control techniques in 2005;111.
HVAC systems. in Energytech, 2012 IEEE, 2012, pp. 1-5. [67] Browne M, Bansal P. Transient simulation of vapour-compression packaged liquid
[32] R. American Society of Heating and E. Air-Conditioning, 2013 ASHRAE Handbook- chillers. Int J Refrig 2002;25:597–610.
18
[68] Kohlenbach P, Ziegler F. A dynamic simulation model for transient absorption HVAC system. Int J Syst Control 2010;1:170–5.
chiller performance. Part II: numerical results and experimental verification. Int J [103] Wang J, Zhang C, Jing Y. Hybrid CMAC-PID Controller in Heating Ventilating and
Refrig 2008;31:226–33. Air-Conditioning System. pp. 3706–3711.
[69] Lei Z, Zaheeruddin M. Dynamic simulation and analysis of a water chiller re- [104] Wang J-J, Zhang C-F, Jing Y-Y. Research of Cascade Control with an Application to
frigeration system. Appl Therm Eng 2005;25:2258–71. Central Air-Conditioning System. pp. 498–503.
[70] Zhang W-J, Zhang C-L, Ding G-L. Transient modeling of an air-cooled chiller with [105] Wang J, Zhang, Jing Y. Analytical design of decoupling control for variable-air-
economized compressor. Part I: model development and validation. Appl Therm volume air-conditioning system. pp. 630–635.
Eng 2009;29:2396–402. [106] Wang J, Jing Y, Zhang C. Robust cascade control system design for central air-
[71] Kim B, Park J. Dynamic simulation of a single-effect ammonia–water absorption conditioning system. pp. 1506–1511.
chiller. Int J Refrig 2007;30:535–45. [107] Trowbridge JT, Ball KS, Peterson JL, Hunn BD, Grasso MM. Evaluation of strate-
[72] Chen Y, Treado S. Development of a simulation platform based on dynamic models gies for controlling humidity in residences in humid climates. ASHRAE Trans
for HVAC control analysis. Energy Build 2014;68(Part A):376–86. 1994:59–73.
[73] Nassif N, Moujaes S, Zaheeruddin M. Self-tuning dynamic models of HVAC system [108] Goyal S, Liao C, Barooah P. Identification of multi-zone building thermal inter-
components. Energy Build 2008;40:1709–20. action model from data. In: Proceedings of the IEEE Conference on Decision and
[74] Mossolly M, Ghali K, Ghaddar N. Optimal control strategy for a multi-zone air Control; 2011, pp. 181–186.
conditioning system using a genetic algorithm. Energy 2009;34:58–66. [109] Myers G, Mitchell J, Nagaoka R. A method of estimating crossflow heat exchanger
[75] Nassif N, Moujaes S, Zaheeruddin M. Self-tuning dynamic models of HVAC system transients. ASHRAE Trans 1965;71:225.
components. Energy Build 2008;40:1709–20. [110] Myers GE, Mitchell JW, Lindeman C. The transient response of heat exchangers
[76] Yao Y, Lian Z, Hou Z. Thermal analysis of cooling coils based on a dynamic model. having an infinite capacitance rate fluid. J Heat Transf 1970;92:269–75.
Appl Therm Eng 2004;24:1037–50. [111] Kabelac S. The transient response of finned crossflow heat exchangers. Int J Heat
[77] Jin G-Y, Cai W-J, Wang Y-W, Yao Y. A simple dynamic model of cooling coil unit. Mass Transf 1989;32:1183–9.
Energy Convers Manag 2006;47:2659–72. [112] Zaheer-Uddin M, Zheng G. A VAV system model for simulation of energy man-
[78] Jin G-Y, Tan P-Y, Ding X-D, Koh T-M. Cooling coil unit dynamic control of in HVAC agement control functions: off normal operation and duty cycling. Energy Convers
system. In: Proceedings of the 2011 6th IEEE Conference on Industrial Electronics Manag 1994;35:917–31.
and Applications; 2011, pp. 942–947. [113] Karkamaz N. A Dynamic Model for Heating and Cooling Coils Order No. 9827730
[79] Wang Y-W, Cai W-B, Li S-J, Xie L-H, Soh Y-C. Development of cooling coil model Ann. Arbor.. University of Colorado at Boulder; 1998.
for system control and optimization. In: Proceedings of the 2002 International [114] Hill J, Jeter S. A linear subgrid cooling and dehumidification coil model with
Conference on Control and Automation, 2002. ICCA. Final Program and Book of emphasis on mass transfer. ASHRAE Trans 1991;97:118–28.
Abstracts; 2002, pp. 133–133. [115] Bocanegra LM, Khan AY. Parametric analysis of heat and mass transfer perfor-
[80] Wang Y-W, Cai W-J, Soh Y-C, Li S-J, Lu L, Xie L. A simplified modeling of cooling mance of a counterflow, cooling and dehumidification coil 1992. New York, NY
coils for control and optimization of HVAC systems. Energy Convers Manag (USA): ASME; 1992.
2004;45:2915–30. [116] Khan A. Heat and mass transfer performance analysis of cooling coils at part-load
[81] Kusiak A, Xu G, Zhang Z. Minimization of energy consumption in HVAC systems operating conditions [54-54]. Trans-Am Soc Heat Refrig Air Cond Eng 1994;100.
with data-driven models and an interior-point method. Energy Convers Manag [54-54].
2014;85:146–53. [117] Wang J, Hihara E. Prediction of air coil performance under partially wet and to-
[82] Atthajariyakul S, Leephakpreeda T. Neural computing thermal comfort index for tally wet cooling conditions using equivalent dry-bulb temperature method. Int J
HVAC systems. Energy Convers Manag 2005;46:2553–65. Refrig 2003;26:293–301.
[83] He X, Zhang Z, Kusiak A. Performance optimization of HVAC systems with com- [118] Brandemuehl MJ, Gabel S, Andresen I, HVAC 2 toolkit: a toolkit for secondary
putational intelligence algorithms. Energy Build 2014;81:371–80. HVAC system energy calculations: prepared for The American Society of Heating,
[84] Aydinalp-Koksal M, Ugursal VI. Comparison of neural network, conditional de- Refrigerating and Air Conditioning Engineers, TC 4.7 Energy Calculations:
mand analysis, and engineering approaches for modeling end-use energy con- ASHRAE; 1993.
sumption in the residential sector. Appl Energy 2008;85:271–96. [119] Kasahara M, Kuzuu Y, Matsuba T, Hashimoto Y. Physical model of an air-condi-
[85] Huang H, Chen L, Mohammadzaheri M, Hu E, Chen M. Multi-zone temperature tioned space for control analysis. ASHRAE Trans 2000;106:304.
prediction in a commercial building using artificial neural network model. In: [120] Ninić N, Vehauc A. The effect of the choice of the enthalpy zero point on cooling
Proceedings of the IEEE International Conference on Control and Automation, tower design and packing data processing. Wärme-und Stoffübertrag
ICCA; 2013, pp. 1896–1901. 1992;27:305–10.
[86] Afram A, Janabi-Sharifi F. Black-box modeling of residential HVAC system and [121] Maiya M. Analysis of modified counter-flow cooling towers. Heat Recovery Syst
comparison of gray-box and black-box modeling methods. Energy Build CHP 1995;15:293–303.
2015;94:121–49. [122] Merkel F, Verdunstungskühlung V. Forchungsarbeiten. ed: Berlin; 1925.
[87] Huang H, Chen L, Hu E. A neural network-based multi-zone modelling approach [123] Poppe M, Rögener H. Berechnung von rückkühlwerken. VDI Wärmeatlas, pp. Mi;
for predictive control system design in commercial buildings. Energy Build 1991.
2015;97:86–97. [124] Bošnjaković F, Technische thermodynamik: T. Steinkopff; 1960.
[88] Jovanović RŽ, Sretenović AA, Živković BD. Ensemble of various neural networks [125] Grange J. Calculating the evaporated water flow in a wet cooling tower. Electricite
for prediction of heating energy consumption. Energy Build 2015;94:189–99. de France (EDF); 1994.
[89] Romero J, Navarro-Esbrí J, Belman-Flores J. A simplified black-box model or- [126] Bourillot C. TEFERI: Numerical model for calculating the performance of an
iented to chilled water temperature control in a variable speed vapour compres- evaporative cooling tower," Electricite de France, 78-Chatou. Thermal Transfer
sion system. Appl Therm Eng 2011;31:329–35. and Aerodynamic Dept; 1983.
[90] Ghiaus C, Chicinas A, Inard C. Grey-box identification of air-handling unit ele- [127] Kloppers JC, Kröger DG. The Lewis factor and its influence on the performance
ments. Control Eng Pract 2007;15:421–33. prediction of wet-cooling towers. Int J Therm Sci 2005;44:879–84.
[91] P. ASHRAE. Heating and cooling, ASHRAE Handbook-HVAC Systems and [128] Stabat P, Marchio D. Simplified model for indirect-contact evaporative cooling-
Equipment, SI ed, American Society of Heating Refrigerating and Air-Conditioning tower behaviour. Appl Energy 8// 2004;78:433–51.
Engineers (ASHRAE) Atlanta, GA, US; 2008. [129] Li X, Li Y, Seem JE. Dynamic modeling of mechanical draft counter-flow wet
[92] Yu FW, Chan KT. Improved energy management of chiller systems by multivariate cooling tower with modelica. ed; 2010, pp. 687–694.
and data envelopment analyses. Appl Energy 2012;92:168–74. [130] Beghi A, Cecchinato L, Paggiaro F, Rampazzo M. VAVAC systems modeling and
[93] Saidur R, Hasanuzzaman M, Mahlia TMI, Rahim NA, Mohammed HA. Chillers simulation for FDD applications. In: Proceedings of the 2011 9th IEEE
energy consumption, energy savings and emission analysis in an institutional International Conference on Control and Automation (ICCA); 2011, pp. 800–805.
buildings. Energy 2011;36:5233–8. [131] Pascual JAF, Garay RS, Torrens JI, Keane MM. Heat ventilation and air con-
[94] Wemhoff AP, Frank MV. Predictions of energy savings in HVAC systems by lumped ditioning modelling for model based fault detection and diagnosis. In: Proc.
models. Energy Build 2010;42:1807–14. BS2013: 13th Conf. Int. Build. Perform. Simul. Assoc., IBPSA, Chambery, France;
[95] Browne MW, Bansal PK. Transient simulation of vapour-compression packaged 2013, p. 351.
liquid chillers. Int J Refrig 2002;25:597–610. [132] Bi Q, Cai W-J, Wang Q-G, Hang C-C, Lee E-L, Sun Y, et al. Advanced controller
[96] Bendapudi S, Braun JE, Groll EA. A comparison of moving-boundary and finite- auto-tuning and its application in HVAC systems. Control Eng Pract
volume formulations for transients in centrifugal chillers. Int J Refrig 2000;8:633–44.
2008;31:1437–52. [133] Huang G, Wang S, Xu X. Robust model predictive control of VAV air-handling
[97] Platt G, Li J, Li R, Poulton G, James G, Wall J. Adaptive HVAC zone modeling for units concerning uncertainties and constraints. HVACR Res 2010;16:15–33.
sustainable buildings. Energy Build 2010;42:412–21. [134] Rehrl J, Horn M. Temperature control for HVAC systems based on exact linear-
[98] Homod RZ, Sahari KSM, Almurib HA, Nagi FH. RLF and TS fuzzy model identifi- ization and model predictive control. In: Proceedings of the 2011 IEEE
cation of indoor thermal comfort based on PMV/PPD. Build Environ International Conference on Control Applications (CCA); 2011, pp. 1119–1124.
2012;49:141–53. [135] Huang G, Wang S, Xu X. A robust model predictive control strategy for improving
[99] J. Wang, C. Zhang, and Y. Jing, "Analytical design of decoupling control for the control performance of air-conditioning systems. Energy Convers Manag
variable-air-volume air-conditioning system," pp. 630-635. 2009;50:2650–8.
[100] Cai J, Kim D, Jaramillo R, Braun JE, Hu J. A general multi-agent control approach [136] Hang C, Chin D. Reduced order process modelling in self-tuning control.
for building energy system optimization. Energy Build 2016;127:337–51. Automatica 1991;27:529–34.
[101] Ghiaus C, Hazyuk I. Calculation of optimal thermal load of intermittently heated [137] Halevi Y. Optimal reduced order models with delay. In: Proceedings of the 30th
buildings. Energy Build 2010;42:1248–58. IEEE Conference on Decision and Control, 1991; 1991, pp. 602–607.
[102] Homod RZ, Sahari KSM, Mohamed HA, Nagi F. Hybrid PID-cascade control for [138] Bi Q, Cai W-J, Lee E-L, Wang Q-G, Hang C-C, Zhang Y. Robust identification of
19
first-order plus dead-time model from step response 1999/01/01 Control Eng [170] Zhao H-x, Magoulès F. A review on the prediction of building energy consumption.
Pract 1999;7:71–7. Renew Sustain Energy Rev 2012;16:3586–92.
[139] Gunay B, Shen W, Newsham G. Inverse blackbox modeling of the heating and [171] Kumar R, Aggarwal R, Sharma J. Energy analysis of a building using artificial
cooling load in office buildings. Energy Build 2017;142:200–10. neural network: a review. Energy Build 2013;65:352–8.
[140] Li Q, Meng Q, Cai J, Yoshino H, Mochida A. Predicting hourly cooling load in the [172] Spindler HC, Norford LK. Naturally ventilated and mixed-mode buildings—Part I:
building: a comparison of support vector machine and different artificial neural thermal modeling. Build Environ 2009;44:736–49.
networks. Energy Convers Manag 2009;50:90–6. [173] Lixing D, Jinhu L, Xuemei L, Lanlan L. Support vector regression and ant colony
[141] Hou ZJ, Lian ZW, Yao Y, Yuan XJ. Cooling-load prediction by the combination of optimization for HVAC cooling load prediction. In: Proceedings of the 2010
rough set theory and an artificial neural-network based on data-fusion technique. International Symposium on Computer, Communication, Control and Automation
Appl Energy 2006;83:1033–46. (3CA); 2010, pp. 537–541.
[142] Kalogirou S, Neocleous C, Schizas C. Building heating load estimation using arti- [174] Zhang Q, Wong YW, Fok SC, Bong TY. Neural-based air-handling unit for indoor
ficial neural networks. In: Proceedings of the 17th international conference on relative humidity and temperature control. ASHRAE Trans 2005;111:63–70.
Parallel architectures and compilation techniques, 1997, p. 14. [175] Ferreira P, Ruano A, Silva S, Conceicao E. Neural networks based predictive
[143] Shin K-W, Lee Y-S. The study on cooling load forecast of an unit building using control for thermal comfort and energy savings in public buildings. Energy Build
neural networks. Int J Air-Cond Refrig 2003;11:170–7. 2012;55:238–51.
[144] Ascione F, Bianco N, De Stasio C, Mauro GM, Vanoli GP. Artificial neural networks [176] Xuemei L, Lixing D, Lanlan L. A novel building cooling load prediction based on
to predict energy performance and retrofit scenarios for any member of a building SVR and SAPSO. In: 2010 International Symposium on Computer,
category: a novel approach. Energy 2017;118:999–1017. Communication, Control and Automation (3CA); 2010, pp. 528-532.
[145] Karatasou S, Santamouris M, Geros V. Modeling and predicting building's energy [177] Gunn SR. Support vector machines for classification and regression. ISIS Tech
use with artificial neural networks: methods and results. Energy Build Report 1998;14.
2006;38:949–58. [178] Dong B, Cao C, Lee SE. Applying support vector machines to predict building
[146] Neto AH, Fiorelli FAS. Comparison between detailed model simulation and arti- energy consumption in tropical region. Energy Build 2005;37:545–53.
ficial neural network for forecasting building energy consumption. Energy Build [179] Zhao HX, Magoules F. Parallel support vector machines applied to the prediction
2008;40:2169–76. of multiple buildings energy consumption. J Algorithms Comput Technol
[147] Yuce B, Li H, Rezgui Y, Petri I, Jayan B, Yang C. Utilizing artificial neural network 2010;4:231–49.
to predict energy consumption and thermal comfort level: an indoor swimming [180] Li Q, Meng Q, Cai J, Yoshino H, Mochida A. Applying support vector machine to
pool case study. Energy Build 2014;80:45–56. predict hourly cooling load in the building. Appl Energy 2009;86:2249–56.
[148] Moon JW, Jung SK, Lee YO, Choi S. Prediction performance of an artificial neural [181] Hou Z, Lian Z. An application of support vector machines in cooling load pre-
network model for the amount of cooling energy consumption in hotel rooms. diction. In: Proceedings of the International Workshop on Intelligent Systems and
Energies 2015;8:8226–43. Applications, 2009. ISA 2009; 2009, pp. 1-4.
[149] Mba L, Meukam P, Kemajou A. Application of artificial neural network for pre- [182] Xuemei L, Lixing D, Lanlan L. A novel building cooling load prediction based on
dicting hourly indoor air temperature and relative humidity in modern building in SVR and SAPSO. pp. 528–532.
humid region. Energy Build 2016;121:32–42. [183] Liang J, Du R. Model-based fault detection and diagnosis of HVAC systems using
[150] Pandey S, Hindoliya D, Mod R. Artificial neural networks for predicting indoor support vector machine method. Int J Refrig 2007;30:1104–14.
temperature using roof passive cooling techniques in buildings in different cli- [184] Pal A, Mudi R. Self-tuning fuzzy PI controller and its application to HVAC systems.
matic conditions. Appl Soft Comput 2012;12:1214–26. Int J Comput Cogn 2008;6:25–30.
[151] Ruano AE, Crispim EM, Conceição EZE, Lúcio MMJR. Prediction of building's [185] Soyguder S, Alli H. Predicting of fan speed for energy saving in HVAC system
temperature using neural networks models. Energy Build 2006;38:682–94. based on adaptive network based fuzzy inference system. Expert Syst Appl 5//
[152] Huang H, Chen L, Mohammadzaheri M, Hu E. A new zone temperature predictive 2009;36:8631–8.
modeling for energy saving in buildings. Procedia Eng 2012:142–51. [186] Calvino F, La Gennusa M, Rizzo G, Scaccianoce G. The control of indoor thermal
[153] Mustafaraj G, Lowry G, Chen J. Prediction of room temperature and relative hu- comfort conditions: introducing a fuzzy adaptive controller. Energy Build
midity by autoregressive linear and nonlinear neural network models for an open 2004;36:97–102.
office. Energy Build 2011;43:1452–60. [187] Chen K, Jiao Y, Lee ES. Fuzzy adaptive networks in thermal comfort. Appl Math
[154] Garnier A, Eynard J, Caussanel M, Grieu S. Low computational cost technique for Lett 2006;19:420–6.
predictive management of thermal comfort in non-residential buildings. J Process [188] Lü H, Jia L, Kong S, Zhang Z. Predictive functional control based on fuzzy T-S
Control 2014;24:750–62. model for HVAC systems temperature control. J Control Theory Appl 2007;5:94–8.
[155] Mechaqrane A, Zouak M. A comparison of linear and neural network ARX models [189] Angelov PP, Filev DP. An approach to online identification of Takagi-Sugeno fuzzy
applied to a prediction of the indoor temperature of a building. Neural Comput models. IEEE Trans Syst Man Cybern Part B (Cybern) 2004;34:484–98.
Appl 2004;13:32–7. [190] Yiu CMJ. Statistical Modelling and Forecasting Schemes for Air-Conditioning
[156] Spindler HC, Norford LK. Naturally ventilated and mixed-mode buildings-Part I: System Order No. 3313066 Ann. Arbor.. Hong Kong Polytechnic University (Hong
thermal modeling. Build Environ 2009;44:736–49. Kong); 2008.
[157] Yokoyama R, Wakui T, Satake R. Prediction of energy demands using neural [191] Ríos-Moreno G, Trejo-Perea M, Castañeda-Miranda R, Hernández-Guzmán V,
network with model identification by global optimization. Energy Convers Manag Herrera-Ruiz G. Modelling temperature in intelligent buildings by means of au-
2009;50:319–27. toregressive models. Autom Constr 2007;16:713–22.
[158] Lu T, Viljanen M. Prediction of indoor temperature and relative humidity using [192] Frausto HU, Pieters JG, Deltour JM. Modelling greenhouse temperature by means
neural network models: model comparison. Neural Comput Appl 2009;18:345–57. of auto regressive models. Biosyst Eng 2003;84:147–57.
[159] Kim W, Jeon Y, Kim Y. Simulation-based optimization of an integrated daylighting [193] Lowry G, Lee M-W. Modelling the passive thermal response of a building using
and HVAC system using the design of experiments method. Appl Energy sparse BMS data. Appl Energy 2004;78:53–62.
2016;162:666–74. [194] Soleimani-Mohseni M, Thomas B, Fahlen P. Estimation of operative temperature in
[160] Chow TT, Zhang GQ, Lin Z, Song CL. Global optimization of absorption chiller buildings using artificial neural networks. Energy Build 2006;38:635–40.
system by genetic algorithm and neural network. Energy Build 2002;34:103–9. [195] Thomas B, Soleimani-Mohseni M. Artificial neural network models for indoor
[161] Tse WL, Chan WL. An automatic data acquisition system for on-line training of temperature prediction: investigations in two buildings. Neural Comput Appl
artificial neural network-based air handling unit modeling. Measurement 2007;16:81–9.
2005;37:39–46. [196] Patil SL, Tantau HJ, Salokhe VM. Modelling of tropical greenhouse temperature by
[162] Moon JW, Jung SK, Kim Y, Han S-H. Comparative study of artificial intelligence- auto regressive and neural network models. Biosyst Eng 2008;99:423–31.
based building thermal control methods – application of fuzzy, adaptive neuro- [197] Seginer I, Boulard T, Bailey BJ. Neural network models of the greenhouse climate.
fuzzy inference system, and artificial neural network. Appl Therm Eng J Agric Eng Res 1994;59:203–16.
2011;31:2422–9. [198] Gouda MM, Danaher S, Underwood CP. Application of an artificial neural network
[163] Ruano A, Crispim E, Conceicao E, Lúcio MM. Prediction of building's temperature for modelling the thermal dynamics of a building's space and its heating system.
using neural networks models. Energy Build 2006;38:682–94. Math Comput Model Dyn Syst 2002;8:333–44.
[164] Hosoz M, Ertunc HM, Bulgurcu H. Performance prediction of a cooling tower using [199] Jiménez MJ, Madsen H, Andersen KK. Identification of the main thermal char-
artificial neural network. Energy Convers Manag 2007;48:1349–59. acteristics of building components using MATLAB. Build Environ 2008;43:170–80.
[165] Swider DJ, Browne MW, Bansal PK, Kecman V. Modelling of vapour-compression [200] Jimenez MJ, Madsen H. Models for describing the thermal characteristics of
liquid chillers with neural networks. Appl Therm Eng 2001;21:311–29. building components. Build Environ 2008;43:152–62.
[166] Asadi E, Silva MG d, Antunes CH, Dias L, Glicksman L. Multi-objective optimiza- [201] Underwood CP. HVAC Control Systems: Modelling, Analysis and Design.
tion for building retrofit: a model using genetic algorithm and artificial neural Routledge; 2002.
network and an application. Energy Build 2014;81:444–56. [202] Moroşan P-D, Bourdais R, Dumur D, Buisson J. Building temperature regulation
[167] Tesiero R, Nassif N, Singh H. Artificial Intelligent Approaches for Modeling and using a distributed model predictive control. Energy Build 2010;42:1445–52.
Optimizing HVAC Systems. In: Proceedings of the 2013 National Conference on [203] Boaventura Cunha J, Couto C, Ruano A. A greenhouse climate multivariable
Advances in Environmental Science and Technology; 2016, pp. 211–220. predictive controller. In: Proceedings of the International Conference and British-
[168] Ferreira PM, Ruano AE, Silva S, Conceição EZE. Neural networks based predictive Israeli Workshop on Greenhouse Techniques towards the 3rd Millennium 534;
control for thermal comfort and energy savings in public buildings. Energy Build 2000, pp. 269–276.
2012;55:238–51. [204] Loveday DL, Craggs C. Stochastic modelling of temperatures for a full-scale oc-
[169] Magnier L, Haghighat F. Multiobjective optimization of building design using cupied building zone subject to natural random influences. Appl Energy
TRNSYS simulations, genetic algorithm, and artificial neural network. Build 1993;45:295–312.
Environ 2010;45:739–46. [205] Frausto HU, Pieters J, Deltour J. Modelling greenhouse temperature by means of
20
auto regressive models. Biosyst Eng 2003;84:147–57. with physical and JIT model for building thermal load prediction and optimal
[206] Patil S, Tantau H, Salokhe V. Modelling of tropical greenhouse temperature by energy saving control. In: Proceedings of the ICCAS-SICE 2009 - ICROS-SICE
auto regressive and neural network models. Biosyst Eng 2008;99:423–31. International Joint Conference 2009; 2009, pp. 2008–2011.
[207] Prívara S, Váňa Z, Gyalistras D, Cigler J, Sagerschnig C, Morari M, et al. Modeling [220] Afram A, Janabi-Sharifi F. Gray-box modeling and validation of residential HVAC
and identification of a large multi-zone office building. In: Proceedings of the 2011 system for control system design. Appl Energy 2015;137:134–50.
IEEE International Conference on Control Applications (CCA); 2011, pp. 55-60. [221] Vaghefi S, Jafari M, Zhu J, Brouwer J, Lu Y. A hybrid physics-based and data
[208] Prívara S, Váňa Z, Cigler J, Ferkl L. Predictive control oriented subspace identi- driven approach to optimal control of building cooling/heating systems. IEEE
fication based on building energy simulation tools. In: Proceedings of the 2012 Trans Autom Sci Eng 2016;13:600–10.
20th Mediterranean Conference on Control & Automation (MED); 2012, pp. [222] Attaran SM, Yusof R, Selamat H. A novel optimization algorithm based on epsilon
1290–1295. constraint-RBF neural network for tuning PID controller in decoupled HVAC
[209] Jiménez MJ, Madsen H. Models for describing the thermal characteristics of system. Appl Therm Eng 2016;99:613–24.
building components. Build Environ 2008;43:152–62. [223] Lü X. Modelling of heat and moisture transfer in buildings: I. Model program.
[210] Yao Y, Huang M, Chen J. State-space model for dynamic behavior of vapor Energy Build 2002;34:1033–43.
compression liquid chiller. Int J Refrig 2013;36:2128–47. [224] Badescu V, Sicre B. Renewable energy for passive house heating: II. Model. Energy
[211] Lennart L. System identification: theory for the user 494. Upper Saddle River, NJ: Build 2003;35:1085–96.
PTR Prentice Hall; 1999. p. 1–14. [225] Badescu V, Staicovici MD. Renewable energy for passive house heating: model of
[212] Kulkarni MR, Hong F. Energy optimal control of a residential space-conditioning the active solar heating system. Energy Build 2006;38:129–41.
system based on sensible heat transfer modeling. Build Environ 2004;39:31–8. [226] Badescu V, Laaser N, Crutescu R. Warm season cooling requirements for passive
[213] Prívara S, Široký J, Ferkl L, Cigler J. Model predictive control of a building heating buildings in Southeastern Europe (Romania). Energy 2010;35:3284–300.
system: the first experience. Energy Build 2011;43:564–72. [227] Scotton F, Huang L, Ahmadi SA, Wahlberg B. Physics-based modeling and iden-
[214] Nishiguchi J, Konda T, Dazai R. Data-driven optimal control for building energy tification for HVAC systems. In: European Control Conference; 2013.
conservation. In: Proceedings of the SICE Annual Conference 2010; 2010, pp. [228] Perera D, Pfeiffer C, Skeie N. Control of temperature and energy consumption in
116–120. buildings-a review. Int J Energy Environ 2014;5:471–84.
[215] Watson I. A case study of maintenance of a commercially fielded case‐based rea- [229] Hou Z, Liu S, Tian T. Lazy-Learning-Based Data-Driven Model-Free Adaptive
soning system. Comput Intell 2001;17:387–98. Predictive Control for a Class of Discrete-Time Nonlinear Systems; 2016.
[216] Tsutsui H, Kamimura K. Chiller condition monitoring using topological case-based [230] Hu C, Jain G, Zhang P, Schmidt C, Gomadam P, Gorka T. Data-driven method
modeling. American Society of Heating, Refrigerating and Air-Conditioning based on particle swarm optimization and k-nearest neighbor regression for esti-
Engineers, Inc., Atlanta, GA (United States) 0001–2505; 1996. mating capacity of lithium-ion battery. Appl Energy 2014;129:49–55.
[217] Sun K, Yan D, Hong T, Guo S. Stochastic modeling of overtime occupancy and its [231] Leephakpreeda T. Grey prediction on indoor comfort temperature for HVAC sys-
application in building energy simulation and calibration. Build Environ tems. Expert Syst Appl 2008;34:2284–9.
2014;79:1–12. [232] McFadden G, Li Z, Rolando Vega PhD P. HVAC load forecasting using LIDAR data
[218] Page J, Robinson D, Morel N, Scartezzini JL. A generalised stochastic model for the and physics-based models. ASHRAE Trans 2015;121. [p. 1DD].
simulation of occupant presence. Energy Build 2008;40:83–98. [233] N/A, "2013 ASHRAE Handbook - Fundamentals (I-P Edition),"ed: American
[219] Iino Y, Murai M, Murayama D, Motoyama I, Kuzusaka S, Ueta K. Hybrid modeling Society of Heating, Refrigerating and Air-Conditioning Engineers, Inc.,
21

Renewable and Sustainable Energy Reviews: Zakia Afroz, GM Sha Fiullah, Tania Urmee, Gary Higgins

Uploaded by

Copyright:

Available Formats

Renewable and Sustainable Energy Reviews: Zakia Afroz, GM Sha Fiullah, Tania Urmee, Gary Higgins

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Renewable and Sustainable Energy Reviews: Zakia Afroz, GM Sha Fiullah, Tania Urmee, Gary Higgins

Uploaded by

Copyright:

Available Formats

Renewable and Sustainable Energy Reviews xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews

Modeling techniques used in building HVAC control systems: A review

1. Introduction control algorithms [6–17]. Improving the system control algorithm,

Nomenclature Te Temperature of the refrigerant at evaporator inlet (°C)

building energy and comfort management. A comprehensive review of

4. Classiﬁcation of HVAC modeling techniques

Sub-Space State Fig. 2. Classiﬁcation of data-driven models used in

4.3. Classiﬁcation of data driven models

Data driven (black box or inverse) modeling technique comprises

The prediction result of this study [95] was found to be within ±

Fig. 3. Schematic of a cross-ﬂow cooling tower [61].

Component of Reference Field of application Method Main Task Outcome Remarks

advanced controllers (e.g., PID, MPC) and energy conservation strate-

Occupants’ schedule can be integrated into the model for

The model suﬀers from one of the drawbacks as stated

It is necessary to validate the model by a real system.

There is a necessity for investigating the predictive

movement of occupants from one zone to another.

holidays can be included in the model for further

Occupants’ overtime schedules on weekends and

improvement in simulation result.

Identiﬁed the model to be used in MPC to control

uncertainty of the weather forecast in its control

occupants’ overtime schedule on building energy

Showed improvement in both building energy

training method were extensively tested for diﬀerent buildings and

the indoor temperature of building.

to train a hybrid/gray box model to predict transient cooling or heating

building energy consumption

Table 3 gives a summary of main surveyed studies on gray box

6. Discussion and summary

The whole structure of HVAC plant is very complex where time-

Improve the accuracy of the building energy

turbances, constraints, and uncertainties are imposed by running the

total HVAC system dynamics. The necessity of appropriate modeling

Survey the importance of considering

comes from the requirement of handling disturbances, constraints, and

Identify the patterns of occupancy

uncertainties existing within the HVAC system dynamics. At the very

the importance of accurate modeling to deal with this highly non-linear

system dynamics has been demonstrated. Next, the function of perfor-

techniques used in HVAC system has been categorized, deﬁned and

critically reviewed to ascertain the application and feature of individual

review. Also, the approach, role, outcome and shortcomings of some of

building energy consumption

the existing studies/developed models where these modeling techni-

Indoor temperature control

Indoor temperature control

strengths and weaknesses of these modeling techniques have been re-

vealed which will be brieﬂy described in this section to perform a

modeling techniques will be compared based on four performance

criteria which are very important for selection of modeling techniques

Garnier, et al. [154]

Prívara, et al. [213]

Kulkarni and Hong

Page, et al. [218]

Sun, et al. [217]