Building and Environment 144 (2018) 159–170

Contents lists available at ScienceDirect

Building and Environment

journal homepage: www.elsevier.com/locate/buildenv

A dynamic gain-scheduled ventilation control system for a subway station T

based on outdoor air quality conditions
Jorge Loy-Benitez1, Qian Li1, Pouya Ifaei, Kijeon Nam, SungKu Heo, Changkyoo Yoo∗
Department of Environmental Science & Engineering, College of Engineering, Kyung Hee University, Yongin, 446-701, South Korea

A R T I C LE I N FO A B S T R A C T

Keywords: Within subway stations, the use of a mechanical ventilation system is a common strategy for improving the
Energy saving indoor air quality (IAQ). These ventilation systems use outdoor air to dilute pollutants on the subway platforms.
Gain scheduling However, a fixed fan speed in manual subway station ventilation systems does not consider variations in IAQ
Indoor air quality dynamics caused by disturbances yielded by the outdoor air quality (OAQ). Since the IAQ in subway stations has
Outdoor air quality
become a major public health concern, this study aims to analyze the IAQ dynamics at different OAQ conditions
Ventilation control system
Ventilation energy
to design a new dynamic ventilation control system. The proposed method implemented a Gain Scheduling
control strategy over OAQ variations at the D-Subway Station in the Seoul metropolitan area. A set of one
feedback (FB) and two feedforward (FF) controllers was implemented. The results showed that the proposed
control system tuned with the internal model control (IMC) method achieved an energy saving of 9% in com-
parison to the manual ventilation system. It was estimated a decrease in energy consumption of 158 kWh/day,
representing an emission reduction of 268 kgCO2/day. Following, an energy cost reduction of 4325 USD yearly
was estimated. Additionally, the indoor particulate matter level is maintained below a control limit considered to
be unhealthy for sensitive groups (150 μg/m3).

1. Introduction
Over the past few decades, underground transportation has become
increasingly important in metropolitan areas worldwide due to its high
efficiency, extensive capacity, traffic relief, and low emissions operation
[1]. Millions of people depend on these transportation systems for their
daily routines, resulting in considerable time spent in subway stations
[2]. However, despite the benefits of subway transport systems, indoor
air pollutants are a potential health risk to commuters [3]. Urban
subway systems are confined, poorly ventilated microenvironments
with unique emission sources (friction in the rail-wheel-brake interac-
tion) of particulate matter (PM) [1,3,4].
Prior studies focused on the monitoring and control of PM generated
in subway stations [2,3,5–11]. The levels of this pollutant in subway
microenvironments are generally higher than at ground level, creating
the potential for hazardous health exposure to commuters [12,13].
Furthermore, PM in subway stations differs from that in other en-
vironments due to shape, distribution, and composition (dominantly
Fe) [4]. Pope and Dockery [14] concluded that long-term PM exposure
is associated with increased cardiovascular mortality. In Seoul, there
are nine lines that are used daily by approximately four million people
[2]. The Korean Ministry of Environment has declared subway stations
as harmful spaces for public health, motivating the establishment of
regulations for the control of these indoor hazardous pollutants [5,6].
For these reasons, this study uses particulate matter smaller than 10 μm
(PM10) to quantify the indoor air quality (IAQ) in the subway station.
IAQ levels in urban subway systems depend on several factors such
as humidity, internal temperature, crowding level and ventilation
conditions [1,6]. In this context, mechanical ventilation systems are
commonly used to improve IAQ by supplying the indoor space with
fresh air from outside [15]. Moreno et al. [3] concluded that subway
platforms are heavily dependent on forced tunnel ventilation to main-
tain low levels of PM. Nonetheless, Kukadia and Palmer [16] conducted
a pilot study concluding that IAQ follows the outdoor air quality (OAQ)
trend to which it is exposed; therefore, IAQ is mainly dependent on
OAQ.
Although a mechanical ventilation system helps in the dilution of
contaminants, it increases the building energy consumption [15,17].
Air conditioning and ventilation systems consume between 35% and
50% of the total electricity in a building space [6,18]. Despite this,
ventilation systems in underground stations are considered to have the
greatest potential for energy savings [19]. The reduction of the demand

∗
Corresponding author.
E-mail address: ckyoo@khu.ac.kr (C. Yoo).
1
The first two authors contributed equally to this paper.

https://doi.org/10.1016/j.buildenv.2018.08.016
Received 22 May 2018; Received in revised form 27 July 2018; Accepted 9 August 2018
Available online 14 August 2018
0360-1323/ © 2018 Elsevier Ltd. All rights reserved.
J. Loy-Benitez et al.
in energy consumption in the building sector has been considered as an
important goal suggested by the Intergovernmental Panel on Climate
Change [18]. Furthermore, some studies have directly related the cur-
rent energy and transportation systems to CO2 discharge over the next
50 years [20]. In South Korea, there are 20 subway lines and 577
subway stations; it is estimated that subway ventilation systems con-
sumption is approximately greater than 1800 KWh/day per station
[5,6]. For these reasons, studies have been conducted to improve IAQ
while saving energy in ventilation systems [2–6,12–14].
Ventilation systems in subway stations work at fixed fan speeds
according to a schedule, regardless of variations in IAQ that manifest as
time-variant disturbance factors. For example, IAQ levels are lowest
during morning rush hours due to overcrowding and heavy use [6].
Additionally, periodic events such as seasonal changes and yellow sand
storms may alter the IAQ conditions [6,21]. Modeling of an IAQ system
comprises all disturbance factors, and it is commonly used as a strategy
to simulate the actual process through mathematical algorithms
[2–5,22].
Previously, to control the IAQ in subway stations, researchers have
carried out several studies using mathematical models. Lee et al. [6]
implemented a gain-scheduled ventilation control system relying on
variations in the IAQ dynamics according to rush and non-rush hours.
Additionally, they proposed a combination of feedback (FB) propor-
tional-integral-derivative (PID), and feedforward (FF) controllers as a
control strategy. The implementation of this system reduced the ven-
tilation energy by 4% maintaining a comfortable IAQ level (PM10 under
120 μg / m3 ). Kim et al. [5] proposed an OAQ-dependent ventilation
system wherein, the amount of PM10 flowing indoors was used as a
Fig. 1. Data measurement in D-Subway station: (a)Tele-monitoring system for IAQ measurements, (b) outdoor ventilators, and (c) location of IAQ measurement
stations.
160
Building and Environment 144 (2018) 159–170
manipulated variable. An FB proportional-integral (PI) and an FF con-
troller were used as a control strategy. This proposed strategy achieved
a reduction in energy consumption of 39% while maintaining a healthy
IAQ level. Vaccarini et al. [23] developed a model-based predictive
control algorithm coupled with a monitoring platform. This study
achieved a greater than 30% energy savings in a Barcelona subway
ventilation system while maintaining comfort levels.
Most of the research considering IAQ has focused on decreasing the
concentration of pollutants in building spaces and on the development
of energy saving strategies. However, considering disturbances as a
control tool for varying IAQ dynamics has not been studied in depth.
The main objective of the present study is to develop a ventilation
control system based on the differentiation of the IAQ system according
to OAQ variations. Hence, IAQ systems and their controllers are up-
dated according to an OAQ signal and specific corrective actions to the
ventilation system are performed for each OAQ condition. As a result,
an energy efficient ventilation system must be generated while main-
taining comfortable IAQ levels.
This paper consists of three major sections. First, the relationships
between indoor PM10 concentration (namely PM10 concentration inside
the subway platform), ventilation system, and disturbances are ana-
lyzed to identify the IAQ system. Second, two IAQ control systems are
defined: one that is derived from the manual ventilation system and the
other from OAQ conditions. In addition, a set of FB and FF controllers
are applied to both systems. In the third section, both ventilation con-
trol systems are simulated, and their performance is compared using the
average indoor PM10 concentration, energy consumption, and occur-
rence of PM10 concentrations exceeding the control limit.
J. Loy-Benitez et al.
2. Materials and methods
2.1. Data description and IAQ system configuration
The dataset used in this study includes in situ measurements from
November 21st to 25th, 2011, in the D-Subway of Seoul metropolitan
city. To keep pollutants at an acceptable and healthy level for com-
muters, the Seoul Metro has implemented a telemonitoring system
(TMS) (Fig. 1). This system measures several parameters of interest
(temperature, humidity, NO, NO2, CO, CO2, PM2.5, and PM10) every
three minutes [2,6]. Fig. 2 shows the variation of each variable in-
volved in the IAQ system expressed as a time-variant series.
The D-subway station consists of two platform levels: 1) the waiting
room, which is the nearest level to the ground at 10 m below the sur-
face, and 2) the underground platform, being the furthest level with
24 m below the surface. The platform is “facing” type, where people
that desire to go to one direction face people that want to go in the
opposite direction, separated by the railway. Operationally, the trains
start running from 5 a.m. to 11 p.m. with an interval of 2–3 min be-
tween them in rush hours (e.g. diurnal, evening), and with an interval
Fig. 2. Measured variables in D-Subway station: (a) Indoor PM10 concentration, (b) ventilation fan speed, (c) outdoor PM10 concentration, and (d) train schedule.
161
Building and Environment 144 (2018) 159–170
less than 10 min in non-rush hours. According to a schedule, the D-
subway station allows the movement of 2000 commuters per hour in
the waiting room and underground platform.
To predict PM10 concentration on the subway platform, an IAQ
system is developed using several variables. The first variable to con-
sider is the concentration of PM10 on the platform (Fig. 2(a)), which
determines the IAQ level in the subway station to which passengers are
exposed. This also denotes the controlled variable on the IAQ system.
The second variable to consider is the fans speed in the ventilation
system (Fig. 2(b)). This variable is denoted as ventilation inverter fre-
quency (Hz) and intends to allow fresh air to enter the station and dilute
the concentration of pollutants. Patterns can be detected for this vari-
able since this speed is fixed for certain hours. From 6 pm to 9 pm, the
fan speed is maintained at its maximum capacity (60 Hz); from 10 p.m.
to 12 a.m., it decreases to 40 Hz; and finally increases to 45 Hz for the
rest of the day. This variable is considered the manipulated variable; as
it is the one that may be regulated to maintain the IAQ at an acceptable
level and defines energy consumption.
Finally, the outdoor PM10 concentration (OPM) and train schedule
(Fig. 2(c) – (d)) are considered as disturbance variables since they upset
J. Loy-Benitez et al. Building and Environment 144 (2018) 159–170

the system and contribute negatively to the IAQ. For this study, the
OPM defines the OAQ, and the Train Schedule variable represents the
frequency of trains arriving at the station, which is assumed as a pro-
portional equivalence of commuters transiting within the subway sta-
tion based on previous studies of Kim et al. [2] and Lee et al. [6].
Fig. 3 illustrates the methodology used in the present study. It is
divided into three phases: 1) IAQ system identification, 2) ventilation
control system design, and 3) ventilation system performance assess-
ment.

2.2. Indoor air quality system identification

As an essential step to control the PM10 concentration in the subway

platform, dynamic models of the PM10 variation should be identified.
Due to the complexity of the IAQ system, an empirical dynamic model
was constructed according to the measured input-output data Fig. 4. Step input test procedure for FOPTD model parameters [26].
[5,6,8,24]. Hence, a First-Order Plus Time-Delay (FOPTD) model was
implemented. This method was selected due to its simplicity when An FOPTD process model is represented by a transfer function given
analyzing the process behavior and determining the target system. in Eq. (1) [26]
Additionally, the behavior of the system is inferred by a simple ob-
y (s ) k exp(−θs )
servation of the model coefficients [5,6,25,26]. G (s ) = =
u (s ) τs + 1 (1)

where y (s ) and u (s ) are the process output and process input respec-
tively, k is the static gain, θ is the time delay, and τ is the time constant.
A step input test method was used for identification of the FOPTD
model for the IAQ system. A schematic procedure for parameters ob-
tained through the step input test is shown in Fig. 4. Additionally, Fig. 5
shows a block diagram that illustrates the IAQ system by integrating the
measured variables.

2.3. Ventilation control design

2.3.1. Feedback control: proportional-integrative-derivative controller

In industrial processes, PID controllers have played a predominant
role due to their easy implementation and good performance. A PID
controller is composed of three parts: proportional Kp , integrative KI ,
and derivative KD . Based on these, there are three common FB con-
trollers utilized, which are the proportional (P), proportional-integral
(PI) and PID. The P control focuses on the reduction of the error signal
to zero. The PI control considers the integral of the error signal over
time, while the PID incorporates a derivative mode to anticipate the
future behavior of the error signal by considering its rate of change. The
structure of a PID controller is described in Eq. (2).
t
kc d (ys (t ) − y (t ))
u (t ) = kc (ys (t ) − y (t )) +
τi
∫ (ys (t ) − y (τ )) dτ + kc τD
dt
0

(2)
Here, ys (t ) , y (t ) , and u (t ) denote the desired process output (set-
point), the actual process output, and the control output of the PID
controller at time t , respectively. PID parameters kc , τi and τD are pro-
portional gain, integral time and derivative time, respectively [25–27].
PID parameters are determined using the dynamic considerations of
the IAQ process model. However, these parameters should be tuned for
high performance. Several tuning methods have been developed in-
cluding internal model control (IMC) and integral of the time-weighted
absolute value of the error (ITAE). It is possible to obtain the para-
meters for each tuning method according to the operations given in
Table 1.

2.3.2. Feedforward control

In a system that is prone to be affected by several disturbances, an
FB controller has disadvantages. FB controllers do not perform the
corrective action until after a deviation in the output occurs, and they
Fig. 3. The general framework of the proposed methodology. do not provide predictive control to compensate for the effects of

162
J. Loy-Benitez et al. Building and Environment 144 (2018) 159–170

Fig. 5. Schematized block diagram of the IAQ system in D-Subway station.

Table 1
Description of PID parameter tuning rules [26].
Controller kc τi τD

IMC (2τ + θ)/2(λ∗ + θ)/ k τ + θ/2 τθ/(θ + 2τ )

ITAE (Set-point) (0.965(θ/ τ )−0.850)/ k τ /((0.796 − 0.1465(θ/ τ )) τ (0.308(θ/ τ )0.929)
ITAE (Disturbance) (1.357(θ/ τ )−0.947)/ k τ /(0.842(θ/ τ )−0.738) τ (0.381(θ/ τ )0.995)

∗λ ≥ 0.25θ .

measurable disturbances. In contrast, an FF controller is used to reject
the effects of disturbances before the process is affected [25,26,28]. The
FF controllers can be mathematically expressed as given in Eq. (3).
uf (s ) Gd (s )
Gff (s ) = − =−
d (s ) Gp (s ) (3)
Here, Gff is the transfer function for the FF controller; uf (s ) and d (s )
are control action and disturbance signal, respectively, and Gd and Gp
are the process models for the disturbance and manipulating variable.
Fig. 6 shows a block diagram of the proposed ventilation control
system, which consists of one FB and two FF controllers. The FB con-
troller controls the indoor PM10 concentration by manipulating the fans
speed (inverter frequency) according to the desired set-point. Ad-
ditionally, the two FF controllers are used to reject the disturbance
effects of train schedule and OPM by means of inverter frequency ma-
nipulation.
2.3.3. Gain Scheduling principle
The Gain Scheduling control strategy incorporates the variation of
the IAQ process dynamics according to the variation of an operating
region [26]. Here, the OAQ is selected as the operating region in terms
of health-risk categorization using the limits established in the com-
prehensive air quality index (CAI) proposed by the Korean Ministry of
Environment. The CAI describes the air quality to which the human
being is exposed in such a way that it can be translated easily into
control actions for the protection of human health [29]. Table 2 sum-
marizes the ranges of the CAI. The low and high breakpoints for each
pollutant are represented by BPLO and BPHI, respectively. These
breakpoint concentrations are used to differentiate between health-risk
categories (good, moderate, unhealthy, and very unhealthy). For the
time interval considered in this study, the OAQ was rated as good or
moderate.
A preliminary analysis of the OPM distribution must be conducted
to select an adequate operating region. Fig. 7(a) shows the monthly
distribution of the OPM within the Seoul metropolitan area in the year
2016 [29]. Here, the red line illustrates the mean value, and the grey
boxes depict each season period. It can be observed that the con-
centration of the pollutant is increased during the spring season (i.e.

Fig. 6. Block diagram of the IAQ ventilation control system in D-Subway station.

163
J. Loy-Benitez et al. Building and Environment 144 (2018) 159–170

Table 2
Comprehensive Air Quality Index (CAI) proposed by the Korean Ministry of Environment [29].
Category Description Good Moderate Unhealthy Very Unhealthy

Values ILO 0 51 101 251

IHI 50 100 250 500

Concentration BPLO BPHI BPLO BPHI BPLO BPHI BPLO BPHI

SO₂ (ppm) 0 0.02 0.021 0.05 0.051 0.15 0.151 1

NO₂ (ppm) 0 0.03 0.031 0.06 0.061 0.2 0.201 2
CO (ppm) 0 2 2.01 9 9.01 15 15.01 50
O₃ (ppm) 0 0.03 0.031 0.09 0.091 0.15 0.151 0.6
PM10 (㎍/㎥) 0 30 31 80 81 150 151 600
PM2.5 (㎍/㎥) 0 15 16 50 51 100 101 500

Note: BPLO and BPHI are the concentration breakpoints for each category description.

March, April, May), when the yellow storm event occurs; then, it gets
lower for the summer season (i.e. June, July, August) and then starts
rising again for the autumn season (i.e. September, October, No-
vember). Following, Fig. 7(b) shows the distribution of OPM on a daily
basis, based on the available measurements from the D-Subway Station.
It should be noted that a specific pattern cannot be identified due to the
stochastic behavior of particulate diffusion because the OPM is influ-
enced by external factors (e.g. seasonality, yellow storm events, heavy
traffic) on a high temporal resolution. The aim of the control strategy is
to generate specific corrective actions based on OAQ conditions;
therefore, it is not convenient to classify an operating region in terms of
a low-resolution temporal basis; instead, the CAI breakpoints are sui-
table candidates to define the operating regions (e.g. good, moderate,
unhealthy, very unhealthy). On the other hand, Fig. 7(c) shows the
distribution of indoor PM10 concentration on the subway platform at
different OAQ health-risk categories. From this analysis, it can be
concluded that the indoor environment remains in better conditions
when outdoor air intake for ventilation occurs at low OPM or good OAQ
(p-value < 0.05). In addition, the outliers of the box plot in Fig. 7(c) are
yielded by the train schedule disturbance, and some other factors that
affect the IAQ and are not considered in this study (e.g. temperature,
humidity, erosion of construction material, etc.) [1,6].
For these reasons, an IAQ system model can be identified for each
OAQ category, as represented in Eq. (4)
μg
k = k1, θ = θ1, τ = τ1; when OPM ≤ 30 (Good OAQ)
m3
μg μg
k = k2, θ = θ2, τ = τ2; when 30 < OPM ≤ 80 3 (Moderate OAQ)
m3 m electrical consumption [6,18,19], energy consumption is selected as the
(4)
2.4. OAQ gain – scheduled ventilation control system proposal with IAQ
dynamic model
The ventilation control system proposed in this paper considers the
variations of the IAQ system dynamics yielded by the changes in OAQ
conditions. To explain how the methods were used, a descriptive
scheme is included in Fig. 8. The methodology has been divided into
three parts: 1) OAQ categorization, 2) gain-scheduled ventilation con-
trol system, and 3) control system performance evaluation. The soft-
ware utilized for the implementation of this methodology was MATLAB
R2016a/Simulink.
First, an incoming signal of OPM is analyzed and categorized ac-
cording to the CAI. Here, the OAQ categories that represent the dif-
ferent IAQ dynamics should be defined. In the second part of the pro-
posed method, a gain-scheduled control system is proposed to capture
the effects of the OPM on IAQ dynamics. For this study, the CAI divides
the OAQ into two categories: 1) good (OPM ≤ 30 μg/m³), and 2)
moderate (30 μg/m³ < OPM ≤ 80 μg/m³). Then, depending on the
category assigned, the IAQ system can be switched. Additionally, an FB
controller and two FF controllers are incorporated. The FB controller is
enhanced using the IMC and ITAE tuning methods for each OAQ ca-
tegory.
Furthermore, the computational implementation of the metho-
dology was developed. First, the data was pretreated using the Rosner's
test for outlier detection on the indoor PM10 concentration, and then,
all variables (i.e. indoor PM10 concentration, inverter frequency, out-
door PM10 concentration and train schedule) are normalized to unit
variance and zero mean. Second, the system identification toolbox was
used to identify the IAQ system, based on the variables mentioned
above. The variables are introduced into the toolbox; then, the FOPTD
models for each OAQ condition are obtained, describing the relation-
ship between the input process variables/manipulated/disturbances
(i.e. inverter frequency, outdoor PM10 concentration, train schedule)
with the output process variable/controlled (i.e. indoor PM10 con-
centration). Third, the FB and FF controllers were calculated for the
IAQ system for good and moderate OAQ (namely, gain-scheduled
ventilation system). Finally, simulations were conducted for each PID
tuning method and arranged to determine energy consumption and the
average indoor PM10 concentration.
The performance of the OAQ gain-scheduled ventilation control
system is evaluated using three indices: 1) energy consumption of the
ventilation system, 2) average PM10 concentration in the platform for
the time interval, and 3) exceeding of indoor PM10 concentration over a
control limit.
Because ventilation systems in subway stations have significant
energy-saving potential and represent a major part of the building's
first index. A third-order polynomial equation (Eq. (5)) is used to esti-
mate the energy consumption of the ventilation system (kWh) as a
function of ventilation inverter frequency (Hz) [9].
Energy Consumption (kWh) = 0.0007 × Hz 3 − 0.046 × Hz 2 + 2.01 × Hz
+ 8.8 (5)
The PM10 concentration in the subway station plays a fundamental
role in the health of commuters [13,15,17]. For this reason, the average
PM10 concentration in the platform is considered as the second index.
For the third index, a comprehensive indoor air quality index (CIAI)
proposed by the U.S. Environmental Protection Agency establishes a
concentration limit less than 150 μg/m³ for PM10 [5], which represents
an “unhealthy for sensitive groups” level of concern. Above this limit,
the adverse effects of PM10 exposure are reflected in personnel working
several hours in the underground environment, children, elderly
people, and those with respiratory and cardiovascular diseases. Con-
sequently, a ventilation system that allows PM10 concentration to in-
crease over the abovementioned breakpoint concentration is considered
to be a risk to human health.
To evaluate the ventilation control system of the proposed method,

164
J. Loy-Benitez et al. Building and Environment 144 (2018) 159–170

Fig. 7. Distribution of (a) seasonal outdoor PM10 concentration in Seoul, (b) daily outdoor PM10 concentration from D-Subway station, and (c) indoor PM10
concentration at different OAQ health-risk categories.

165
J. Loy-Benitez et al.
Fig. 8. Illustration of the proposed method for the gain-scheduled ventilation control system in D-Subway station.
the OAQ gain-scheduled ventilation control system is compared with
the manual ventilation system and the controlled ventilation system
(without Gain Scheduling) using the aforementioned indices.
3. Results and discussion
A Gain Scheduling control strategy was applied to the problem of
the IAQ dynamics variation according to OAQ conditions. Data from D-
Subway Station were used as a control, and the results are detailed in
the following sections.
3.1. Manual IAQ system identification
The capture of the IAQ system dynamics is described mathemati-
cally by FOPTD transfer functions for the prediction of indoor PM10
concentration. The transfer functions describe the relationships be-
tween PM10 concentration in the platform and ventilation fan speed,
train schedule, and OPM based on the manual ventilation system. The
following equation shows the transfer function of PM10 concentration
on the platform due to the manual ventilation system.
−1.2127 exp( −0.063s )
Gp =
0.0168s + 1
According to Eq. (6), a negative value for the static gain indicates
that the PM10 concentration of the platform decreases 1.2127 times
when ventilation fan speed increases. The time delay from action by the
ventilation system to a change in PM10 concentration on the platform is
0.063 days (91 min). Similarly, it takes 0.017 days (25 min) for the
platform PM10 concentration to reach 63% of its maximum value.
The influences of the train schedule and OPM on the dynamics of
indoor PM10 concentration are described by Eqs. (7) and (8).
2.9783 exp( −0.0319s )
Gd1 =
0.0223s + 1
0.5261
Gd2 =
0.0537s + 1
The static gains for the train schedule and OPM transfer functions
show positive values. Therefore, the disturbances cause an increase in
the indoor PM10 concentration up to 2.9783 and 0.5261 times for every
unit disturbance signal change of train schedule and OPM, respectively.
The time needed for the disturbances to affect the indoor PM10
concentration is 0.0319 days (46 min) and 0 days for the train schedule
and the OPM, respectively. Thus, it is concluded that the OPM im-
mediately affects the IAQ system, increasing the PM10 concentration on
the platform.
The time constants are 0.0223 days (32 min) and 0.0537 days
(77 min) for OPM and train schedule, respectively. These constants
Building and Environment 144 (2018) 159–170
describe the time needed after each disturbance to reach 63% of the
maximum indoor PM10 concentration.
3.2. OAQ gain-scheduled process modeling
The Gain Scheduling control in this study differentiates the dy-
namics of the IAQ system depending on an OAQ categorization.
Therefore, two IAQ systems are identified and expressed as transfer
functions for good and moderate OAQ. Equations (9) and (10) show the
IAQ system differentiated by good and moderate OAQ categories, re-
spectively.
−0.9758 exp( −0.0524s )
GGood OAQ =
0.0261s + 1 (9)
−1.6701 exp( −0.0387s )
G Moderate OAQ =
0.0222s + 1 (10)
According to the equations, both cases indicate that an increased
signal on the ventilation system may decrease the indoor PM10 con-
centration. The times required to cause a change in the indoor PM10
concentration are 0.0524 days (75 min) and 0.0387 days (56 min) for
good and moderate OAQ, respectively. In comparison to the manual
(6) IAQ system, the time delay constants for both gain-scheduled IAQ
systems are reduced; similarly, the time constants decrease. Since the
indoor PM10 concentration in good OAQ condition is lower than when
the OAQ is moderate, the ventilation system dilutes the indoor pollu-
tant concentration almost immediately. Nevertheless, more time is re-
quired to return the indoor PM10 concentration to lower levels. In ad-
dition, the absolute value of the static gain for the IAQ system at good
OAQ conditions is lower than that of the other systems.
3.3. PID tuning parameters for gain-scheduled systems
(7) The PID controllers' parameters for the manual IAQ system and the
proposed OAQ gain scheduled were calculated through IMC and ITAE
methods as shown in Table 3. The parameters of the PID controller for
(8)
IMC and ITAE tend to be negative, which implies the direction of the
ventilation control action. For example, when the indoor PM10 con-
centration surpasses the setpoint, a negative input action will correct
the response until reaching the setpoint.
The absolute value of Kp in Table 3 defines the control efforts in this
type of controllers. Additionally, it affects the integral and derivative
parts. The efforts to satisfy the setpoint are greater when the IAQ system
remains in the good OAQ category. Thus, the inverter frequency must
make a greater effort to achieve the dilution of pollutants within the
platform using outdoor air in good conditions.
The value of KI is obtained by dividing the proportional gain by the
integral time. As the absolute value of KI approaches zero, the
166
J. Loy-Benitez et al. Building and Environment 144 (2018) 159–170

Table 3 though the absolute value of KD for the tuning methods converges to
PID parameters tuned with the IMC and ITAE methods. zero (Table 3), the derivative parameters tuned by ITAE show smaller
Tuning Tuning parameters values compared to the values obtained from the IMC method. There-
method fore, the ITAE method shows less stable control than the other tuning
Proportional gain Integral gain Derivative gain method.
(K p = k c ) (KI = k c / τi ) (KD = k c τd )

IMC Manual −0.506 −10.4712 −0.00555

Good −0.8177 −15.6344 −0.01071
3.4. Ventilation control system
Moderate −0.0523 −12.3869 −0.00529
To control the IAQ in the D-Subway Station, a manual ventilation
ITAE Manual −0.259 −3.8088 −0.00466 system is used. However, a fixed rate for the inverter frequency has
Good −0.546 −10.5 −0.00819
been established according to a schedule detailed in section 2.1. During
Moderate 0.36 −8.7805 −0.00396
rush hours, the inverter frequency is maintained at its maximum ca-
pacity, 60 Hz. Meanwhile, during non-rush hours, the inverter fre-
Table 4 quency operates in the 40–45 Hz range. Nonetheless, the manual ven-
Ventilation system characteristics of D-Subway station. tilation system cannot properly handle the PM10 concentration on the
platform; due to the variations in IAQ according to OAQ conditions.
Feature Value Unit
Therefore, a control system for manual ventilation of the D-Subway
Ventilation unit 2 n Station (namely, controlled ventilation system) is implemented for
Static pressure 105 mmHg maintaining indoor PM10 concentration at healthy levels by the com-
Interval of inverter frequency 0–60 Hz
bination of a PID controller and two FF controllers (Fig. 7). Then, dif-
Ventilation capacity 60,000 m3/h
Filter efficiency 0.8 – ferent IAQ dynamics according to OAQ conditions are incorporated by
Air flow per inverter frequency 1000 m3/Hz the Gain Scheduling principle (namely, OAQ gain-scheduled ventilation
system). The ventilation control systems are operationally conditioned
by the ventilation system properties (e.g. inverter frequency, ventila-
controlled variable returns to the setpoint very slowly. In contrast, a tion capacity, filtration capacity) in the D-Subway Station given in
high value of KI may result in an unstable and poorly controlled system Table 4.
for the indoor PM10 concentration due to a fluctuation in the inverter The ventilation system consists of 2 ventilation units with an overall
frequency. For both tuning rules, the absolute value of KI is higher capacity of 60,000 m3/h, the captured outside air is then passed
when the control system is within the good OAQ region (Table 3). The through a fabric panel filter to finally supply this air to the platform.
parameters of the integral part computed by the ITAE method may The inverter frequency operability ranges from 0 to 60 Hz. However, it
allow the control system to remain more stable than in the other has been suggested that the minimum fan speed needs to supply fresh
method. air to the platform even if the IAQ is good. Therefore, the minimum
The KD tends to stabilize the controlled process. A large value of KD value for the inverter frequency is set at 20 Hz. Simulations were con-
reduces the settling time improving the stability of a process. Even ducted using a 3-min time step and a fixed setpoint of 70 μg/m3.

Fig. 9. Performance comparison between manual ventilation system and controlled ventilation system tuned with (a) IMC and (b) ITAE tuning methods (fixed set-
point: 70 μg/m3).

167
J. Loy-Benitez et al.
Fig. 10. Control performance comparison between controlled ventilation systems and OAQ gain-scheduled systems tuned with (a) IMC and (b) ITAE tuning methods
(fixed set-point: 70 μg/m3).
Finally, performance comparisons were made in terms of the tuning
method used in the FB controller.
The results of the manual ventilation system and the controlled
ventilation system are shown in Fig. 9. In comparison to the manual
ventilation system, the controlled ventilation system does not show
large fluctuations in indoor PM10 concentration. Additionally, during
the ventilation period, the indoor PM10 concentration remains below
the control limit of 150 μg/m3. During non-operational hours (12
a.m.–6 a.m.), the controlled ventilation system maintains the inverter
frequency at low values. Then, for the disturbance patterns of rush
hours (7 a.m.–10 a.m. and 6 p.m.–10 p.m.) and OPM, the inverter
frequency increases progressively to prevent an increase of PM10 con-
centration on the platform. Therefore, the controlled ventilation system
keeps indoor PM10 concentration at healthy levels while decreasing
energy consumption.
A comparison between the controlled ventilation system and the
OAQ gain-scheduled ventilation system is shown in Fig. 10. Regarding
the inverter frequency, the OAQ gain-scheduled ventilation control
system tuned with IMC and ITAE methods generates lower values
compared to the controlled ventilation system. This decrease is evi-
denced during rush hours, while the PM10 concentration maintains the
same variation for both cases. It is notable that, for one rush hour in-
terval (night rush hour of day 4), the indoor PM10 concentration is
increased. In this case, the PM10 concentration is caused by high dis-
turbance effects of train schedule and some other factors not considered
in this study. The OAQ gain-scheduled ventilation system reflects that a
maximum fan speed in deteriorated OAQ conditions may affect the IAQ.
From another perspective, this system considers that the dilution of
contaminants in the platform is more efficient with an outside air intake
in good conditions.
Control performances of the manual ventilation system, controlled
ventilation system, and OAQ gain-scheduled ventilation control system
are shown in Table 5. In terms of energy consumption, the OAQ gain-
scheduled ventilation system saves the most energy in comparison to
168
Building and Environment 144 (2018) 159–170
the manual ventilation system and the controlled ventilation system.
The best performance regarding energy consumption is given by the
OAQ gain-scheduled ventilation control system tuned with the IMC
tuning method. However, indoor PM10 concentration is increased due
to fixed parameters in the PID controllers that are used at a large time
delay. Thus, for rush hours, the decrease in PM10 concentration on the
platform is conflicting. To tackle these antagonistic issues, further in-
vestigations must be carried out on the dynamic recalculation of the
PID parameters.
For further analysis, Fig. 11 shows a comparison between the
manual ventilation system and the OAQ gain-scheduled ventilation
system (IMC tuned). There is an increase in the PM10 concentration on
the platform and a decrease in the ventilation inverter frequency. It
should be noted that the proposed control strategy explores the IAQ
dynamics and detects the non-operational hours. Hence, the proposed
system allows the fans to be set at their lowest speed to keep the subway
platform ventilated on the non-operational hours. Consequently, an
increase in the PM10 concentration during these hours is likely to occur;
however, the risk to public health gets lower than operational hours due
to the low quantity of commuters; generally, subway workers transit
within the subway station during these hours. An important contribu-
tion of this study arises to provide valuable information to let em-
ployers evaluate and make decisions on the safety and protection of
workers during non-operational hours. This final trade-off between
inverter frequency and PM10 concentration represents a saving on un-
necessary air delivery from the ventilation system and therefore, both
control systems show a considerable energy saving potential. Ad-
ditionally, it is notable that both control systems presented in this study
(i.e. controlled ventilation system and OAQ gain-scheduled ventilation
system) save energy rather than decreasing the indoor PM10 con-
centration. This is because, in non-operational hours, the significant
effect of rush hours disturbance gets much lower than in operational
hours; then, the fan speed is set to the minimum capacity (20 Hz), al-
lowing a deterioration on the IAQ while saving energy. Nonetheless,
J. Loy-Benitez et al.
Table 5
Control performance evaluation based on average PM10 concentration on the platform, energy consumption and exceeding occurrence with a control limit con-
centration.
Tuning method Average PM10 conc. on the platform (μg/m3)
Manual ventilation system 64
Controlled manual ventilation system IMC 69.95
ITAE 69.84
OAQ gain scheduled ventilation system IMC 70.13
ITAE 70.35
due to the low transit of commuters during non-operational hours, the
exposure of a larger quantity of commuters to the deteriorate IAQ
ambient is not significant.
Through the implementation of the OAQ gain-scheduled ventilation
control system tuned with the IMC method, excessive intervals in en-
ergy consumption were reduced. Considering a ratio of 1.7 kg CO2/
kWh, energy consumption represents a fundamental impact in terms of
greenhouse gas (GHG) emissions and contributes to anthropogenic
climate change [20,30]. The OAQ gain-scheduled ventilation control
system tuned with the IMC method produces a decrease of 158 kWh per
day in comparison with the conventional system. Therefore, CO2
emissions due to energy consumption may be reduced by 268 kgCO2
per day and 97,820 kg CO2 per year. Another factor to consider is the
economic saving in energy costs. The cost of energy in South Korea is
0.075 USD/kWh [6]. Thus, the reduction in energy consumption would
represent a daily saving of 11.87 USD and up to 4325 USD yearly.
Annually, implementation of the proposed control strategy in all
subway stations in the Seoul metropolitan area (20 lines, 577 subway
stations) could reduce up to 56,442,140 kg CO2 emission and 2,495,669
USD in energy costs. Therefore, it is concluded that the proposed
method is an economic solution to strengthen the energy saving po-
tential of ventilation systems in subway stations.
4. Conclusions and future work
A novel ventilation control system has been developed for a subway
station in the metropolitan area of Seoul, through the application of a
Gain Scheduling control strategy. This method considers the changes in
IAQ system dynamics according to OAQ conditions.
The strategy is complemented by a set of one FB controller and two
FF controllers. Then, the results were expressed in terms of the tuning
Fig. 11. Trade-off comparison between manual ventilation system and OAQ gain-scheduled ventilation system (IMC tuned).
169
Building and Environment 144 (2018) 159–170
Energy consumption (kWh/day) Exceeding points PM10 ≥ 150 μg/m3
1811 6
1729 0
1726 0
1653 0
1662 0
method used for the FB controller: IMC and ITAE. Finally, the best
control performance was obtained when the OAQ gain-scheduled con-
trol ventilation system was tuned with the IMC method, wherein energy
consumption was reduced by 9%, and a healthy IAQ level (indoor PM10
concentration below 150 μg/m3) was obtained. A Gain Scheduling
control strategy provides corrective actions depending on time-varying
operating regions. Therefore, this method could be applied to ventila-
tion systems in public use infrastructures and industrial applications in
which the ventilation system is a primary factor in productivity such as
underground mining, tunneling, and chemical processing.
Nonetheless, it has been noted some limitation in the development
of this study, which can be detailed as follows:
• Despite the benefits of the proposed ventilation control system, this
study presents a limitation due to the lack of consideration of other
outdoor pollutants (e.g., CO2, NO2, etc.) on the IAQ and on the
ventilation system performance, which has not been studied in
depth.
• The conflicting trade-off between the indoor PM10 concentration
and the inverter frequency of the ventilation system may be tackled
by changing the fixed setpoint (70 μg/m3) to a real-time setpoint
recalculation.
• To control the IAQ on the subway platform, a continuous scheduled
response was found. However, the stochastic behavior of the OPM
does not allow scheduling of the control system for the inverter
frequency on a future time-interval basis.
•
Due to the lack of open source data, a holistic model for capturing
the variation of indoor PM10 concentration due to OAQ conditions
cannot be made. To tackle this issue, access to a complete dataset
needs to be granted.
J. Loy-Benitez et al.
As a future work, we consider the necessity of further investigations
to tackle the issues abovementioned; principally, the implementation of
a predictive model able to capture the OPM temporal sequence in-
tegrated into the inverter frequency of the ventilation system. The ap-
plication of deep learning predictive models could be useful for the
control schedule in subway stations. It is expected that the contribution
of these future studies will support the development of smart systems in
buildings in terms of energy efficiency and human health.
Acknowledgments
This work was supported by the National Research Foundation of
Korea (NRF) grant funded by the Korea government (MSIT). (No. NRF-
2017R1E1A1A03070713).
Appendix A. Supplementary data
Supplementary data related to this article can be found at https://
doi.org/10.1016/j.buildenv.2018.08.016.
References
[1] V. Martins, T. Moreno, L. Mendes, K. Eleftheriadis, E. Diapouli, C.A. Alves,
M.C. Minguillón, Factors controlling air quality in different European subway sys-
tems, Environ. Res. 146 (2016) 35–46 http://doi.org/10.1016/j.envres.2015.12.
007.
[2] M. Kim, B. Sankararao, O. Kang, J. Kim, C. Yoo, Monitoring and prediction of in-
door air quality (IAQ) in subway or metro systems using season dependent models,
Energy Build. 46 (2012) 48–55 http://doi.org/10.1016/j.enbuild.2011.10.047.
[3] T. Moreno, N. Pérez, C. Reche, V. Martins, E. de Miguel, M. Capdevila, ...
W. Gibbons, Subway platform air quality. Assessing the influences of tunnel ven-
tilation, train piston effect and station design, Atmos. Environ. 92 (2014) 461–468
https://doi.org/10.1016/j.atmosenv.2014.04.043.
[4] E. Guo, H. Shen, L. He, J. Zhang, Investigation of air pollution of Shanghai subway
stations in ventilation seasons in terms of PM2.5 and PM10, Toxicol. Ind. Health 33
(7) (2017) 588–600 http://doi.org/10.1177/0748233717693547.
[5] M. Kim, R.D. Braatz, J.T. Kim, C. Yoo, Indoor air quality control for improving
passenger health in subway platforms using an outdoor air quality dependent
ventilation system, Build. Environ. 92 (2015) 407–417 http://doi.org/10.1016/j.
buildenv.2015.05.010.
[6] S. Lee, S. Hwangbo, J.T. Kim, C.K. Yoo, Gain scheduling based ventilation control
with varying periodic indoor air quality (IAQ) dynamics for healthy IAQ and energy
savings, Energy Build. 153 (2017) 275–286 http://doi.org/10.1016/j.enbuild.2017.
08.021.
[7] Y.S. Kim, M.J. Kim, J.J. Lim, J.T. Kim, C.K. Yoo, Predictive monitoring and diag-
nosis of periodic air pollution in a subway station, J. Hazard Mater. 183 (1–3)
(2010) 448–459 http://doi.org/10.1016/j.jhazmat.2010.07.045.
[8] M.H. Kim, Y.S. Kim, J. Lim, J.T. Kim, S.W. Sung, C.K. Yoo, Data-driven prediction
model of indoor air quality in an underground space, Kor. J. Chem. Eng. 27 (6)
(2010) 1675–1680 http://doi.org/10.1007/s11814-010-0313-5.
[9] H. Liu, S. Lee, M. Kim, H. Shi, J.T. Kim, K.L. Wasewar, C. Yoo, Multi-objective
optimization of indoor air quality control and energy consumption minimization in
a subway ventilation system, Energy Build. 66 (2013) 553–561 http://doi.org/10.
1016/j.enbuild.2013.07.066.
[10] G.Q. Li, S.J. You, A new system to reduce air pollution in metro platform, Procedia
170
Building and Environment 144 (2018) 159–170
Environmental Sciences 11 (PART C) (2011) 1454–1458 http://doi.org/10.1016/j.
proenv.2011.12.218.
[11] S. Sadokierski, Design of a modern subway ventilation system, Tunnels Tunn. Int.
36 (11) (2004) 48–50 http://doi.org/10.13140/RG.2.1.1322.6968.
[12] L. Guo, Y. Hu, Q. Hu, J. Lin, C. Li, J. Chen, ... H. Fu, Characteristics and chemical
compositions of particulate matter collected at the selected metro stations of
Shanghai, China, Sci. Total Environ. 496 (2014) 443–452 https://doi.org/10.1016/
j.scitotenv.2014.07.055.
[13] J. Han, S. bark Kwon, C. Chun, Indoor environment and passengers' comfort in
subway stations in Seoul, Build. Environ. 104 (2016) 221–231 https://doi.org/10.
1016/j.buildenv.2016.05.008.
[14] C.A. Pope, D.W. Dockery, Health effects of fine particulate air pollution: lines that
connect, J. Air Waste Manag. Assoc. 56 (6) (2006) 709–742 https://doi.org/10.
1080/10473289.2006.10464485.
[15] T. Paul, D. Sree, H. Aglan, Effect of mechanically induced ventilation on the indoor
air quality of building envelopes, Energy Build. 42 (3) (2010) 326–332 https://doi.
org/10.1016/j.enbuild.2009.09.010.
[16] V. Kukadia, J. Palmer, The effect of external atmospheric pollution on indoor air
quality: a pilot study, Energy Build. 27 (3) (1998) 223–230 http://doi.org/10.
1016/S0378-7788(97)00044-3.
[17] T. Paul, D. Sree, H. Aglan, Effect of mechanically induced ventilation on the indoor
air quality of building envelopes, Energy Build. 42 (3) (2010) 326–332 http://doi.
org/10.1016/j.enbuild.2009.09.010.
[18] M.J. Kim, R.D. Braatz, J.T. Kim, C.K. Yoo, Economical control of indoor air quality
in underground metro station using an iterative dynamic programming-based
ventilation system, Indoor Built Environ. 25 (6) (2016) 949–961 https://doi.org/
10.1177/1420326X15591640.
[19] Z. Yang, Z. Yu, L. Yu, F. Ma, Research on frequency conversion technology of metro
station's ventilation and air-conditioning system, Appl. Therm. Eng. 69 (1–2) (2014)
123–129 https://doi.org/10.1016/j.applthermaleng.2014.04.016.
[20] K.B. Debnath, M. Mourshed, Forecasting methods in energy planning models,
Renew. Sustain. Energy Rev. 88 (September 2017) (2018) 297–325 https://doi.org/
10.1016/j.rser.2018.02.002.
[21] Y.S. Chung, M.B. Yoon, On the occurrence of yellow sand and atmospheric loadings,
Atmos. Environ. 30 (13) (1996) 2387–2397 http://doi.org/10.1016/1352-
2310(95)00251-0.
[22] S. Lee, H. Liu, M. Kim, J.T. Kim, C. Yoo, Online monitoring and interpretation of
periodic diurnal and seasonal variations of indoor air pollutants in a subway station
using parallel factor analysis (PARAFAC), Energy Build. 68 (PARTA) (2014) 87–98
https://doi.org/10.1016/j.enbuild.2013.09.022.
[23] M. Vaccarini, A. Giretti, L.C. Tolve, M. Casals, Model predictive energy control of
ventilation for underground stations, Energy Build. 116 (2016) 326–340 https://
doi.org/10.1016/j.enbuild.2016.01.020.
[24] R.K. Calay, W.C. Wang, A hybrid energy efficient building ventilation system, Appl.
Therm. Eng. 57 (1–2) (2013) 7–13 https://doi.org/10.1016/j.applthermaleng.
2013.03.042.
[25] D.E. Seborg, D.A. Mellichamp, T.F. Edgar, F.J. Doyle III, Process Dynamics and
Control, John Wiley & Sons, 2010.
[26] S.W. Sung, I.-B. Lee, Process Identification and PID Control, John Wiley & Sons,
2009.
[27] A. Roy, K. Iqbal, PID controller tuning for the first-order-plus-dead-time process
model via Hermite-Biehler theorem, ISA (Instrum. Soc. Am.) Trans. 44 (3) (2005)
363–378 http://doi.org/10.1016/S0019-0578(07)60210-9.
[28] L. Ljung, Prediction error estimation methods, Circ. Syst. Signal Process. 21 (1)
(2002) 11–21 https://doi.org/10.1007/BF01211648.
[29] Air Korea: Real Time Air Quality. Introduction to the CAI, https://www.airkorea.or.
kr/eng/cai; [accessed 10 January 2018].
[30] D. Weisser, A guide to life-cycle greenhouse gas (GHG) emissions from electric
supply technologies, Energy 32 (9) (2007) 1543–1559 http://doi.org/10.1016/j.
energy.2007.01.008.