Building and Environment 72 (2014) 332e342

Contents lists available at ScienceDirect

Building and Environment

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

Sensor-aware elevator scheduling for smart building environments

Ohhoon Kwon a, Eunji Lee a, Hyokyung Bahn b, *
a
School of Electrical and Computer Engineering, Seoul National University, Seoul 151-742, South Korea
b
Smart Building Convergence Research Center, Ewha University, Seoul 120-750, South Korea

a r t i c l e i n f o a b s t r a c t

Article history: In a modern smart building, sensors can detect various physical information such as temperature, hu-
Received 24 May 2013 midity, motion, and light, which can be used for smart living services. This paper presents an efficient
Received in revised form elevator scheduling system by making use of indoor sensor technologies. Specifically, multiple sensor
9 November 2013
devices consisting of RFID, video, and floor sensors are used together to detect the candidate elevator
Accepted 13 November 2013
passengers’ behavior before they push the elevator call button. The detected information is then deliv-
ered to the elevator scheduling system through building networks. By using this information, our
Keywords:
scheduling system generates a reservation call for candidate passengers and controls the moving di-
Elevator scheduling
Sensor
rection and the moving time of elevator cars efficiently. Extensive simulations with various passengers’
Smart building traffic conditions show that the proposed system performs significantly better than the conventional
Smart living elevator scheduling system in terms of the average waiting time, the maximum waiting time, and the
energy consumption of elevator systems.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction control system can recognize the arrival of passengers before

Recent advances in sensor technologies enable a plenty of smart
living services in our daily lives. For example, in a smart building
environment, various physical information such as temperature,
humidity, motion, light, and sound can be collected from sensors,
and then transferred immediately to the building control system.
This allows the detection and tracking of environmental conditions
and human behaviors, which will be exploited for a plenty of ser-
vices such as energy-saving, comfort, healthcare, and security [1].
HVAC (Heating, Ventilating, and Air Conditioning) is a representa-
tive system that utilizes sensor technologies. HVAC controls the air
condition and temperature of a building according to the weather
and the existence of human beings in the building.
Elevator scheduling is another important service that can be
involved in sensor technologies in a smart building. There is a
report that HVAC and elevator services are the two major com-
plaints of building tenants [2]. We aim to improve this situation
by presenting a novel elevator scheduling system that utilizes
indoor sensor technologies. As human behaviors and movements
can be estimated precisely with current sensor technologies such
as camera, audio, optical, and floor sensors [1,3e5], an elevator
* Corresponding author. Smart Building Convergence Research Center, Ewha
University, 11-1 Daehyun-dong, Seodaemun-gu, Seoul 120-750, South Korea.
Tel.: þ82 2 3277 2368; fax: þ82 2 3277 2306.
E-mail addresses: ohkwon@oslab.snu.ac.kr (O. Kwon), ejlee@oslab.snu.ac.kr
(E. Lee), bahn@ewha.ac.kr (H. Bahn).
actual call buttons are pushed. The system, then, utilizes this
information for an efficient scheduling of elevator cars, leading to
reduced latency and power consumption. Though lots of re-
searches on elevator scheduling have been performed to achieve
the same goal, utilizing sensor technologies to obtain passenger
information is in the initial stage of elevator scheduling prob-
lems. To the best of our knowledge, our previous research was
the first attempt to do so but it was limited to a single sensor
technology and is not applicable to group elevator systems [6].
This paper uses multiple sensor technologies to predict passen-
gers’ information more precisely. We validate the effectiveness of
the prediction system and present a general framework for the
group elevator scheduling system that uses these multiple sensor
technologies.
There are several performance criteria of elevator scheduling
systems such as minimizing the waiting time, the riding time, and
the energy consumption. Most existing studies have focused on
minimizing the average waiting time since passenger’s dissatis-
faction grows rapidly as the waiting time increases [7]. Our elevator
scheduling system focuses on reducing the energy consumption as
well as the average waiting time.
The elevator scheduling problem is difficult because of compli-
cated elevator dynamics, uncertain traffic in various patterns, and
multiple goals to be optimized simultaneously. Thus, previous
studies use various optimization techniques such as genetic algo-
rithms and fuzzy systems [8e10,11e28]. Also, to collect more

0360-1323/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.buildenv.2013.11.013
 O. Kwon et al. / Building and Environment 72 (2014) 332e342 333

information on waiting passengers, some facilities such as cameras
and additional buttons are adopted [10,18]. Some research predicts
passengers’ traffic pattern using peak-time distribution to consider
human life patterns [9,13,14]. However, prediction and adaptation
used in existing research is limited as elevator scheduling systems
recognize passengers’ information only after they push the elevator
call button. Unlike previous studies, our system detects passengers’
information by smart sensor technologies before they arrive at the
elevator, and uses this detected information in scheduling elevator
cars efficiently. Specifically, we generate a reservation call for
candidate passengers detected by sensors and control the
moving direction and the moving time of elevator cars for the
reservation calls.
To evaluate the performance of our elevator scheduling system,
we conduct experiments with various workload conditions. The
results show that the proposed elevator scheduling system per-
forms better than the conventional elevator scheduling system in
terms of the passengers’ average waiting time, the maximum
waiting time, and the energy consumption of the elevator system
significantly. Specifically, the improvement of the average waiting
time and the energy consumption is in the range of 15e30% and
28e31%, respectively, by controlling the moving time and the
moving direction of elevator cars for reservation calls.
The remainder of this paper is organized as follows. In Section 2,
we provide previous studies that have been proposed for elevator
scheduling systems and sensor technologies for smart building
environments. In Section 3, we describe our proposed elevator
scheduling system in detail. Section 4 shows the performance
evaluation results of the proposed system. Finally, we conclude the
paper in Section 5.
2. Related works
2.1. Elevator scheduling systems
There have been a variety of studies on the scheduling of
elevator systems to reduce passengers’ waiting time. Most of them
focus on the group elevator systems, in which a single control unit
schedules multiple elevators simultaneously. In group elevator
systems, the scheduling problem becomes even more complicated,
and thus various optimization techniques such as fuzzy systems
[12,22e24,26,29], genetic algorithms [10,15,21], genetic network
programming [11], artificial neural networks [13,27,28], and DNA
computing methods [42] have been adopted.
Igarashi et al. present a fuzzy system based group elevator
scheduling systems [12]. When a passenger pushes the elevator call
button, the system evaluates each elevator car using a fuzzy func-
tion and assigns the car with the largest evaluation value that can
minimize the waiting time of the request. Similar studies have also
been performed by Kaneko et al. [22]. Unlike Igarashi et al., Kaneko
et al. consider the riding time of the elevator as well.
Kim et al. use genetic algorithms to optimize the scheduling of
group elevator systems [10]. They employ cameras to detect the
number of waiting passengers at each floor and assign multiple

Table 1
A summary of elevator scheduling systems.

System proposers Heuristics used Optimization goals System overhead Main research issue Physical characteristics

Gudwin et al. [23] Fuzzy systems with Reducing waiting time Relatively high Dynamic scheduling e
linear context adaptation with fuzzy systems
Igarashi et al. [12] Fuzzy systems Reducing waiting time Relatively high Scheduling with fuzzy systems e
Kurosawa and Hill climbing and Reducing waiting time Relatively high Scheduling with e
Hirasawa [24] fuzzy systems combined optimization
Kaneko et al. [22] Fuzzy systems Reducing waiting time Relatively high Scheduling with fuzzy systems e
Ishikawa et al. [26] Fuzzy systems Reducing waiting time Relatively high Scheduling with fuzzy systems e
and riding time
So et al. [27] Neural networks Reducing waiting time Relatively high Traffic pattern recognition e
with neural networks
Dewen et al. [13] Neural networks Reducing waiting time Relatively high Scheduling with e
neural networks
Markon et al. [28] Adaptive control with Reducing waiting time Relatively high Scheduling adaptively e
neural networks with neural networks
Fujino et al. [15,21] Floor attribute control Reducing waiting time, Relatively high Scheduling with GA e
with genetic algorithm riding time, and considering floor attributes
car crowding
Eguchi et al. [11] Genetic network Reducing waiting time Relatively high Scheduling with genetic e
programming and riding time network programming
Chan et al. [20] Servicing different floors Reducing waiting time Low Scheduling specific floors e
with predetermined cars
Kim et al. [29] Fuzzy systems Reducing waiting time Relatively high Traffic classified and e
and energy consumption fuzzy-based scheduling
Kim et al. [10] Genetic algorithms, Reducing waiting time, Relatively high GA-based scheduling Camera at
multiple car load balancing with camera each floor
assignment with cameras
Lee et al. [8] Linear algorithm Reducing waiting time Low Scheduling based on e
and energy consumption waiting time and energy
Amano et al. [18] Replacing up-down buttons Reducing waiting time Low Scheduling with Numeric buttons
by numeric buttons at each floor a new user interface at each floor
Brand et al. [9] Dynamic programming under Reducing waiting time Low Up-peak/down-peak e
Marcov decision process analysis and parking strategy
Pepyne and Dynamic threshold based Reducing waiting time Low Up-peak traffic e
Cassandras [14,16] policy under Marcov model analysis and scheduling
Strang and Bauer [40] Context-aware algorithm Reducing waiting time Low Context-awareness techniques RFID tags
and riding time
Zhao and Liu [42] DNA computing Reducing waiting time Relatively high Reducing candidate e
schedules through DNA computing
Zhang and Zong [43] Robust optimization model Reducing waiting time Relatively high Up-peak traffic analysis e
and energy consumption and reducing energy
334 O. Kwon et al. / Building and Environment 72 (2014) 332e342
Fig. 1. Tracking by video sensors in the aware home [3]. In this prototype home, the location system knows user’s behaviors: where to go and when to go.
elevators simultaneously if the number of waiting passengers ex-
ceeds the capacity of a single elevator car.
Some recent studies formulate the elevator scheduling problem
as an optimization problem in large search spaces and reduce the
spaces through approximation techniques. Hiller and Klug present
the re-optimization technique that computes new schedules each
time a new passenger arrives and reduces the searching space
through branch-and-bound techniques [44]. Zhao and Liu model
the elevator scheduling problem as an optimization problem in a
large search space and reduce the search space by using DNA
computing methods [42].
If an elevator scheduling system knows the destination floor of
the passengers before they ride the car, the system could schedule
elevators more efficiently. A conventional elevator system, how-
ever, has only two buttons (upward and downward) at each floor,
and thus it cannot predict passengers’ destination in advance. To
address this problem, Amano et al. present an elevator system that
has the destination-floor buttons at each floor. When a passenger
intends to ride an elevator, he/she pushes a specific destination-
floor button at the floor. This provides more information to the
elevator control system than just up and down buttons [18]. Dai
et al. use hybrid genetic algorithms to solve the group elevator
scheduling problems with similar interfaces [30]. Strang and Bauer
use RFID tags to inform the destination floor before passengers
enter the elevator car [40]. They also present the elevator sched-
uling system that switches its scheduling algorithm according to
context information. Kwon et al. also use context information by
using floor sensors, but their scheduling is not applicable to group
elevator systems [6].
Mitsubishi Electric Corporation developed an RFID-enabled
elevator system [31]. The RFID tag informs the arrival of passen-
gers and their destination, which are used for the elevator control
and security. By combining RFID and cameras, their system discerns
Fig. 2. User tracking systems with floor sensors [4,5,33].
if a person wants to use an elevator or is just walking near an
elevator [31].
In most elevator environments, traffic patterns change peri-
odically within a day, and a number of studies have presented
traffic pattern based elevator scheduling systems. Pepyne and
Cassandras analyze up-peak traffic patterns of elevator systems
and exploit the pattern in the elevator scheduling [14,16]. Brand
et al. consider both up-peak and down-peak traffics [9]. Their
system dispatches an empty elevator to a desired parking loca-
tion before requests from users. The location is determined by
the analysis of up-peak and down-peak traffic patterns. These
parking strategies could minimize passengers’ waiting time in
common cases. The traffic of an elevator system changes
continuously as time progresses. There have been researches on
detecting and modeling these traffic changes. Dewen et al. pre-
sent a learning paradigm for elevator scheduling systems [13].
Their system detects the current flow and chooses a scheduling
policy among a set of ready-prepared policies. Pepyne et al.
present an elevator control strategy that delays the elevator
moving until the number of passengers inside it reaches a certain
threshold [14].
Some recent studies consider the energy consumption of the
elevator system as well as the passengers’ waiting time [8,29,34].
Since reducing the energy consumption and the passengers’
waiting time sometimes conflict, optimization of these multiple
goals becomes an even more complex problem. Kim et al. use fuzzy
systems to solve this problem [29]. Lee et al. use a weighted sum
technique to satisfy the two goals and periodically adjust the
weight value [8]. Zhang and Zong observe that most energy is
consumed under up-peak traffic patterns and present an energy-
saving scheduling optimization for up-peak patterns [43]. Table 1
summarizes various elevator scheduling systems and their
characteristics.
 O. Kwon et al. / Building and Environment 72 (2014) 332e342
Hall/Car Call List Moving with passengers
P1 (t4, 1, 11) Moving without passengers
P2 (t12, 16, 1) ti: j Car is at jth floor when time is ti
P3 (t20, 18, 1)
Px (ti, j, k) Passenger x wants to move
from jth to kth floor at ti
t51: 18
t19: 16 P3
P2
t14: 11
P1
t34: 1 t68: 1
t4: 1
Fig. 3. An example of the conventional elevator scheduling system.
2.2. Sensor technologies for smart building environments
Recently, smart building prototypes are developed that contain
various indoor sensors such as video sensors, floor sensors, and
small battery-powered wireless devices. We can obtain plenty of
context information from these sensors and the information can be
exploited for smart home services. Kidd et al. present the aware
home prototype [3]. It consists of several living spaces such as
bedroom, bathroom, living room, dining room, kitchen, laundry
room, and office. In this prototype room, video sensors are used to
trace human locations. A person in this home is tracked and his
activities are reported on the map. With this information, human
behaviors can be analyzed and predicted. Fig. 1 shows a person
tracked from the video sensor in the aware home.
Orr et al. present the smart floor for user identification and
tracking [4]. They have created a system for identifying people
according to their footstep force profile. Because people interact
with the smart floor system, the location of the people can be
traced and also be predicted. Addlesee et al. present the active floor,
which is similar to the smart floor [5]. The active floor is the square
grid of conventional tiles that is supported at the corners by cy-
lindrical load cells which send weight changes of about 50 g to the
location system. With this information, the system can predict the
future locations of people. Similarly, Steinhage and Lauterbach
present an array of capacitive sensors embedded in floor carpets to
track the movement across a large area [33]. Fig. 2 shows these floor
sensors.
Want et al. present the active badge location system for the
tracking of people in an office environment [32]. In this system,
people wear an active badge, and it periodically transmits signals.
These periodic signals are picked up by the network of sensors and
used to trace the location of people.
Sensor networks have also been widely used for environmental
monitoring and controlling situations appropriately. Gao et al.
propose an intelligent lighting control system based on a wireless
sensor network [45]. Their system reduces the energy consumption
of a smart building significantly and provides comfortable building
environment. Eliades et al. present the sensor placement problem
in intelligent buildings for monitoring and protecting indoor air
335
quality against contamination events [46]. They formulate the
problem as a multi-objective optimization problem to minimize the
sensor cost as well as the impact of contamination events. Sreed-
haran et al. design a real-time sensor system that combines infor-
mation from multiple sensors to accurately detect the sudden
release of toxic contaminants in a smart building [47]. Chen and
Wen present a new sensor system design methodology based on an
indoor airflow model to protect building environments from
dangerous contaminants [48]. Liu and Zhai present the optimal
sensor network design to protect a building from critical indoor
contamination [49].
Smart parking providers also make use of sensor technologies to
monitor the real-time availability of urban parking spaces [36].
Finding the best path to an exit with sensor networks is also
attempted [37]. Brunette et al. developed wireless sensor nodes
that can provide contextual information including human activity,
environment, and RFID tags, which can be used in ubiquitous
computing applications to automatically adjust their behavior to
the situation [41]. Ramirez et al. suggest a technology that detects
forest fire by making use of wireless sensor networks and fusion
information [38]. Robertson et al. present a scheme that detects
unusual incidents or counts cars with inductive loop sensors and
controls traffic intelligently [39].
3. The proposed system
In the elevator scheduling system, there are two types of calls.
One is the hall call and the other is the car call [29,35]. A hall call is
issued when a passenger pushes up or down button at each floor. A
car call is issued when a passenger pushes a destination-floor
button inside the elevator car. Conventional elevator scheduling
systems could not recognize passengers before they actually push
the hall call or car call buttons. Due to such limited scheduling in-
formation, passengers’ waiting time and energy consumption
highly increases in some cases. Take, for example, a typical situation
where a passenger pushes a hall call button at the 1st floor
immediately after the elevator moves from the 1st to the top floor.
Hall/Car Call List Moving with passengers
P1 (t1, t4, 1, 11) Moving without passengers
P2 (t9, t12, 16, 1) ti: j Car is at jth floor when time is ti
P3 (t17, t20, 18, 1) Px (ti, tj, k, l) Passenger x wants to move
From kth to lth floor at tj
Reservation call occurs at ti
t21: 18
P3
t23: 16 P2
t14: 11
P1
t4: 1 t38: 1
Fig. 4. An example of an efficient elevator scheduling system with the reservation call
that is issued by the passenger information obtained from sensors.
336 O. Kwon et al. / Building and Environment 72 (2014) 332e342

Sensor-aware elevator scheduling system

Assignment Unit

Control
Unit

Reservation Unit

Hall calls Sensing information

Fig. 5. Basic architecture of the sensor-aware elevator scheduling system.

Then, the passenger should wait during a round trip time of the car,
and energy consumption of the elevator system also increases due
to the long moving distance of the car.
To address such problems, we propose a novel elevator sched-
uling system called SES (Sensor-aware Elevator Scheduling). SES
obtains passengers’ information in advance from multiple sensors
and exploits the information to schedule elevator systems. As
mentioned in Section 2, with rapid advances in sensor and wireless
network technologies, there is no difficulty in detecting and
transferring the passenger information in advance. Given infor-
mation on subsequent passengers in advance, SES schedules
elevator cars more efficiently in terms of the waiting time and
energy consumption.
Figs 3 and 4 show an example that compares the conven-
tional elevator scheduling system and SES. In this example, there
are three passengers, P1, P2, and P3, waiting for the elevator car.
P1 is at the 1st floor and aims to go up to the 11th floor. P2 and
P3 are at the 16th and 18th floors, respectively, and intend to go

Moving with passengers

Hall/Car Call List
Moving without passengers
P1 (t1, t31, 18, 1)
ti: j
P2 (t7, t37, 6, 18) Car is at jth floor when time is ti
Px (ti, tj, k, l) Passenger x wants to move
From kth to lth floor at tj
Reservation call occurs at ti
t18: 18 t31: 18 t65: 18 t49: 18

P1 P1

P2
P2
t53: 6 t19: 6
t37: 6

t1: 1 t48: 1 t14: 1 t66: 1

Non-delayed movement Delayed movement

Fig. 6. An example of non-delayed movement and delayed movement.

 O. Kwon et al. / Building and Environment 72 (2014) 332e342
Hall/Car Call List
P1 (t1, t31, 12, 15)
ti: j
P2 (t6, t36, 6, 18)
Px (ti, tj, k, l) Passenger x wants to move
t59: 18
t38: 15
t35: 12 P1
t31: 8 P2
t47: 6
Conventional scheduling
Fig. 7. An example of the moving direction (MD) value used in the sensor-aware elevator scheduling system.
down to the 1st floor. Passengers P1, P2, and P3 push the hall call
button at t4, t12, and t20, respectively. In this simple example, we
use the logical time that increases by one whenever the car
moves upward or downward for a floor and ignore all other
times such as the boarding time. The car first starts from the 1st
floor with P1 and goes up to the 11th floor. After P1 gets out at
the 11th floor, the car continues to go up to the 16th floor to pick
up P2. When the time becomes t19, the car starts to go down
from the 16th floor with P2. Unfortunately, P3 pushes the hall
call button at the 18th floor immediately after the car has left the
16th floor. If the elevator scheduling system knows the hall call
at the 18th floor before the car starts to move down, the car may
move up to the 18th floor first to pick up P3; then P3 waits only
for one time unit. This could eventually reduce the average
waiting time of the three passengers significantly. However, it
was not possible as conventional elevator systems could not
know the time of passengers’ requests before they actually push
the hall call button.
Fig. 4 shows an efficient elevator scheduling with this infor-
mation for the same example. If sensors detect the candidate pas-
sengers’ requests three time units before they actually arrive in
front of the elevator door, the elevator scheduling system recog-
nizes the three passengers’ requests when the time is t1, t9, and t17,
respectively. With this information, the car moves from the 1st
floor to the 11th floor with P1, and then goes up to the 16th floor to
pick up P2. While going to the 16th floor, the scheduling system
recognizes that P3 wants to go down. Hence, instead of stopping at
the 16th floor, the car passes by the 16th floor, and then goes
straight up to the 18th floor to pick up P3. P2 takes the car at time
t23 after the car goes down from the 18th floor. When compared to
the conventional elevator scheduling system, P2 should wait 4
more time units but the waiting time of P3 is dramatically reduced.
Furthermore, the total running time of the car is reduced by 30
time units. Due to the significant decrease in the running time of
the car, energy consumption is also reduced. In a smart building
337
Moving with passengers
Moving without passengers
Car is at jth floor when time is ti
From kth to lth floor at tj
Reservation call occurs at ti
t48: 18
t45: 15
t42: 12 P1
t34: 8
P2
t36: 6
Sensor-aware scheduling
environment, therefore, we can minimize passengers’ average
waiting time and energy consumption by using the information
from sensors.
3.1. Sensor-aware elevator scheduling (SES)
In this section, we explain the details of the proposed sensor-
aware elevator scheduling (SES) system. Fig. 5 shows the overall
architecture of SES; it consists of three components, namely the
Control Unit (CU), the Reservation Unit (RU), and the Assignment
Unit (AU). CU controls the moving or stopping of the elevator car.
RU collects and processes passenger information detected by sen-
sors. When a passenger approaches the elevator, multiple sensors
detect the passenger’s location in advance, and then send this in-
formation to the elevator control system through the building
network. RU calculates the time when the passenger will arrive at
the elevator and then generates a reservation call according to the
calculated time. AU receives the reservation call from RU, and
makes two decision values; Delay Time (DT) and Moving Direction
(MD). DT is the time interval that should be delayed before the car
starts its movement for the reserved request, and MD is the moving
direction, i.e., upward or downward, of the car. If a reservation call
arrives during the idle time of the elevator car, the car does not
move immediately at right, but delays its movement as long as it
can reach the requested location before the passenger arrives there.
Determining the DT value is important to reduce the moving dis-
tance of the car, potentially related to the energy consumption,
since other requests towards the same direction during the delayed
time can arrive. We define DT as the time difference between the
passenger’s moving time to the elevator door (denoted by PT) and
the elevator car’s moving time to the reserved floor. In particular,
the elevator car’s moving time depends on the distance of the
movement. If the distance is longer than a certain threshold
(denoted by MIN), a uniform motion interval exists in the middle of
the moving. Otherwise, the car moves only as acceleration and
338 O. Kwon et al. / Building and Environment 72 (2014) 332e342

Fig. 8. Algorithm of the sensor-aware elevator scheduling system.

deceleration motions. Equation (1) represents how the DT value is actual hall call time, and RCT is the reservation call time. Using
calculated. Expression (1), we can estimate the time when the car starts to
move.
8 qffiffiffiffiffiffiffiffi
> Now, we will explain our scheduling system by describing the
>
< PT  2 DIST
ASC if DIST < MIN
workings of the algorithm with appropriate examples. Fig. 6 shows
DT ¼
>
>  qffiffiffiffiffiffiffi  an example of the delayed movement. There are two passengers P1
: PT  2 MIN þ DISTMIN otherwise and P2. P1 is at the 18th floor and aims to move down to the 1st
ASC USC (1)
floor. P2 is at the 6th floor and aims to move up to the 18th floor. P1
PT ¼ AHT  RCT and P2 make reservations at t1 and t7, and they will be in front of the
DIST ¼ jCF  PFi j$FH elevator at t31 and t37, respectively. When we do not apply the
delayed movement, the car starts moving at t1 towards the 18th
where CF is the current floor of the car, PFi is the current floor of floor to pick up P1. Although P2 makes a reservation at t7, the car
passenger i, FH is the height of a floor, DIST is the difference in already passed by the 6th floor at that time. As a result, P2 should
height between the car’s and the passenger i’s floors, MIN is the wait for a long time, and also the system spends additional energy
minimum distance for uniform motion, ASC is the uniform accel- due to the long moving distance. When we use delayed movement,
eration of the car, USC is the uniform velocity of the car, AHT is the however, the car delays its movement for DT (13 time units in this
 O. Kwon et al. / Building and Environment 72 (2014) 332e342 339

PWT ðx; i; jÞ ¼ TMOVE þ TADDITIONAL

8 qffiffiffiffiffiffiffiffi
>
> 2 DIST if DIST < MIN
< ASC
TMOVE ¼ qffiffiffiffiffiffiffi (2)
>
>
: 2 MIN þ DISTMIN otherwise
ASC USC

TADDITIONAL ¼ TOPEN þ TBOARDING þ TCLOSE

Fig. 9. Prototype experiments to collect sensor data and compare them with actual
calls.
example), and then starts moving at t14; as a result, P2 can take the
car at this turn.
Moving direction (MD) is also important to minimize pas-
sengers’ waiting time. Fig. 7 depicts an example for the effect of
MD. Suppose that reservation calls for passengers P1 and P2 are
issued at t1 and t6, from the 12th and 6th floors, respectively,
and the passengers will actually arrive in front of the elevator
door 30 time units later; the elevator car initially stops at the
8th floor and the destination of P1 and P2 are the 15th and 18th
floors, respectively. Then, in the conventional elevator sched-
uling system, the car first moves to the 12th floor at t31 to pick
up P1. P1 rides in the car at t35, and reaches the destination
floor at t38. Then, the car moves to the 6th floor to pick up P2. In
this way, the waiting time of P2 becomes 11 time units. On the
contrary, in the sensor-aware elevator scheduling system, the
car first goes down to the 6th floor to pick up P2, and then
moves to the 12th floor to pick up P1. Accordingly, the average
waiting time of the passengers and the moving distance of the
where DIST is the distance between the passenger and the car, MIN
is the minimum distance for uniform motion, ASC is the uniform
acceleration of the car, and USC is the uniform velocity of the car.
TOPEN, TBOARDING, and TCLOSE represent the time to open the elevator
door, board the elevator, and close the elevator door, respectively. If
the car should visit some additional floor during the movement
from i to j, multiple TMOVE and TADDITIONAL values are added to PWT.
After the allocation phase is completed, SES-G inserts the pas-
senger’s reservation call into the request queue of the selected car.
SES-G, then, processes the passenger’s reservation call based on the
original SES algorithm.
4. Performance analysis
4.1. Detecting passenger information through sensors
To assess the effectiveness of multiple sensor devices in
detecting passenger information, we equipped RFID, video, and
floor sensors in a twenty-story building for residence, and collected
sensed data for enrolled users during a day. The building authen-
ticates the enrolled users at the gate of the underground parking
area or the main entrance of the ground floor using an RFID tag or a
password. We first collect sensor data at this time to make a
reservation call. Less than 120 s are needed for a candidate pas-
senger to move from this location to the elevator door. Along the
25

car are significantly reduced. Fig. 8 shows the algorithm of the CS

Average waiting time (sec)

proposed SES. 20 SES-G-30

SES-G-60
15 SES-G-120
3.2. Extending to group elevator scheduling
10
In a building with group elevator systems, it is difficult to control
multiple elevator cars efficiently. In this section, we describe the 5
extension of the original SES for group elevator scheduling systems
called SES-G (SES for Group elevators). The operation of SES-G 0
consists of two phases: the allocation phase and the processing 5 10 15 20 25 30 35

phase. When a reservation call is issued, the elevator scheduling Arrival rate (passengers/min.)
system first decides which car should be allocated to that request. (a) Average waiting time
For efficient allocation, SES-G calculates the passengers’ expected 3000
waiting time for each car. Then SES-G allocates the car incurring the CS
Energy consumption (KJ)

minimum expected waiting time to that request. The expected 2500

SES-G-30
waiting time (PWT) of elevator car x from i-th to j-th floor is SES-G-60
2000
computed as SES-G-120
1500

Table 2 1000
Experimental conditions.
500
Items Settings
0
Number of floors 20 5 10 15 20 25 30 35
Number of cars 6
Arrival rate (passengers/min.)
Elevator capacity 20 people/car
Moving speed 3 m/s
Floor height 3m
(b) Energy consumption
Open/close time 2s
Fig. 10. Performance evaluation results for the conventional scheduling system (CS)
Boarding time 2s
and the sensor-aware elevator scheduling system (SES) as the arrival rate varies.
340 O. Kwon et al. / Building and Environment 72 (2014) 332e342

hallway towards the elevator, floor sensors are located, and they average arrival rate ranging from 5 to 35 passengers/min as usual
can detect the moving direction of people through the path of the [8,10,29]. To reflect the peak-time traffic in real situations, we also
step. This information is collected about 60 s prior to the actual hall generate passengers’ traffic by a non-homogeneous Poisson process
call at the elevator. Video sensors are also located at the ceiling of where the arrival rate itself changes according to another Poisson
the hallway, and they can also recognize the movement of people. process [8,10]. We use the following three criteria to compare SES
The distance from this location to the elevator is about 30 s or less. with CS (Conventional elevator Scheduling system).
We set the default moving direction of reservation calls to “up” at
the parking area and the ground floor. Note that a reservation call is  Average waiting time: a waiting time is the interval between a
dropped from the request queue when it is not continued to appear passenger’s hall call time and a car’s arrival time. Average
in the subsequent sensors. For example, when a reservation call is waiting time is the average of all passengers’ waiting time.
issued 120 s prior to the actual hall call through RFID tags but there  Maximum waiting time: the maximum waiting time is the largest
is no corresponding reservation call through video or floor sensor, passenger’s waiting time during the experimental period [11].
the request is dropped from the scheduling queue.  Energy consumption: the energy consumption is the total
In the residence area ranging from the 2nd to 20th floors, each amount of energy consumed during the operation of the
home has its front door and there are floor sensors along the elevator system [8].
hallway. 60 s or so are needed to move from the front door to the
elevator including the locking time of the door. Video sensors are According to the prototype settings in Section 4.1, we perform
also located at the ceiling of the hallway, and the distance to the three SES configurations: SES-G-30, SES-G-60, and SES-G-120. In
elevator is about 30 s or less. We set the default moving direction of SES-G-30, the reservation call is issued 30 s before passengers make
reservation calls to “down” at residence floors. the actual hall call. Similarly, SES-G-60 and SES-G-120 issue the
We gathered the sensor data during a day and compared the reservation calls 60 s and 120 s before actual hall calls, respectively.
reservation calls issued by sensor information and actual hall calls In reality, the optimum time period for the reservation call should
from passengers. We use a CCTV in front of the elevator car to be defined empirically depending on passengers’ and elevators’
collect the actual hall call information. The comparison result moving time and the location of sensors. An elevator’s moving time
showed that 84.3% of the reservation calls eventually invoke actual can be estimated precisely by the law of motion as in Equation (1),
calls. As reservation calls are dropped from the request queue if but a passenger’s moving time varies depending on the location of
subsequent actual calls do not appear, the effect of false reservation sensors and each passenger’s step. If we overestimate the passen-
calls is not significant. Note that our scheduler waits only for a ger’s moving time, the passenger should wait long for the elevator
certain limited time for reservation calls, and it promptly reacts to car. If we underestimate the passenger’s moving time, the elevator
the changed request information. On the other hand, most requests car may arrive before the passenger arrives. As a closer sensor es-
benefit from these reservation calls by shortening their waiting timates the moving time of the candidate passenger more precisely,
time. Fig. 9 depicts our experimental environments that collect in our empirical study, a reservation call from a distant sensor is
passengers’ information in advance with sensing system. dropped from the request queue when it appears from a closer
sensor. To set an appropriate time period for reservation calls, we
4.2. Experimental results use the median value of the passengers’ moving time to the
elevator door captured in the traces. Location of sensors is also
In this section, we present the performance evaluation important to determine the time period for reservation calls. A
results for SES. Table 2 shows the experimental conditions for distant sensor provides the scheduling system with passengers’
performance evaluation. We use similar conditions with previous information earlier and thus using this information allows the
studies [8,10]. The number of floors and elevators is set to 20 and 6, elevator car to pick up many passengers at a time. This eventually
respectively, and each elevator car can accommodate 20 people. leads to the reduction of energy consumption. In contrast, a close
The passengers’ traffic is generated by Poisson process with the sensor provides more accurate information, and thus the average

20 44 3000
19
Energy consumption (KJ)
Maximum waiting time (sec)
Average waiting time (sec)

42
18 2500

17 40
2000
16
38
15 1500
36
14
13 1000
34
12
32 500
11
10 30 0
CS

CS
SES-G-30

SES-G-60

SES-G-30

SES-G-60
SES-G-120

SES-G-120

CS

SES-G-60
SES-G-30

SES-G-120

(a) Average waiting time (b) Maximum waiting time (c) Energy consumption
Fig. 11. Performance evaluation results for the conventional scheduling system (CS) and the sensor-aware elevator scheduling system (SES) as the traffic is non-homogeneous.
 O. Kwon et al. / Building and Environment 72 (2014) 332e342
waiting time of passengers can be minimized. For now, we show
the results for a spectrum of policies by utilizing multiple sensors
and do not provide the tuning issue of the optimal time period. We
think that much research is needed towards this direction.
Fig. 10(a) shows the average waiting time of CS and SES groups
(SES-G-30, SES-G-60 and SES-G-120) as the arrival rate varies. As
shown in the figure, SES-G performs consistently better than CS
irrespective of the arrival rate. Specifically, the average waiting time
of SES-G-30 is better than CS by 22.1% on average and up to 30.0%.
The performance gain of SES-G-120 is relatively small because the
car picks up many passengers at a time. Fig. 10(b) shows the total
amount of energy consumed during the operation of the elevator
system. SES groups again perform better than CS. Among the three
configurations, SES-G-120 performs the best as the car picks up
more people at a single round trip time than other elevator
scheduling systems. The performance improvement of SES-G-120
against CS is as much as 28.6% on average and up to 31.2% in
terms of the energy consumption.
Fig. 11 shows the results for non-homogeneous traffic patterns.
Again, SES-G performs better than CS for all performance measures.
As shown in Fig. 11(a), the average waiting time of SES-G-30 is 16.1%
better than CS. Similar to the results in Fig. 10, the performance gain
of SES-120 is relatively small in terms of the average waiting time.
However, this is not the case for the maximum waiting time as
depicted in Fig. 11(b). Irrespective of SES configurations, the
improvement of the maximum waiting time is as much as 6.5e
6.8%. Fig. 11(c) shows the total amount of energy consumed during
the operation of the elevator system. Similar to the homogeneous
traffic cases, SES-120 performs the best; the improvement is 30.7%
compared to CS in terms of the energy consumption.
5. Conclusion
In this paper, we presented a novel elevator scheduling system,
called SES, for smart buildings. SES uses multiple sensor devices to
detect passengers’ information before they actually push the
elevator call button. Specifically, three types of sensor devices, RFID,
video, and floor sensors are used to estimate passenger information
precisely. By utilizing this prior information, SES controls the
elevator system more efficiently than conventional on-demand
elevator schedulers. We generate a reservation call for candidate
passengers detected by sensors and control the moving direction
and the moving time of elevator cars for the reservation calls.
Experimental results with various traffic patterns showed that SES
reduces the average waiting time and the energy consumption by
15e30% and 28e31%, respectively, compared to legacy elevator
systems.
As a future work, we plan to extend SES to collect the destina-
tion floor of passengers in advance by using a variety of sensors
including smartphone sensors. We expect that more complete in-
formation can be used for efficient scheduling. Another direction of
future research is to design a scheduler considering the change of
power rates. Due to recent advances in smart grid technologies, the
price of the electricity changes dynamically within a day. We aim to
design an elevator scheduling system that adapts the change of
power rates to reduce electricity charges. For example, when the
power rate is low, the scheduler controls elevators in the normal
mode, while it switches to a low-power mode when the power rate
becomes high.
Acknowledgment
This work was supported by the National Research Foundation
of Korea (NRF) grant funded by the Korea government (MEST) (No.
2011-0028825).
341
References
[1] Essa I. Ubiquitous sensing for smart and aware environments. IEEE Pers
Commun 2000;7:47e9.
[2] So AT, Liu SK. An overall review of advanced elevator technologies. Elev World
1996;44:96e103.
[3] Kidd CD, Orr RJ, Abowd GD, Atkeson CG, Essa IA, MacIntyre B, et al. The aware
home: a living laboratory for ubiquitous computing research. Lect Notes
Comput Sci 1999;1670:191e8.
[4] Orr RJ, Abowd GD. The smart floor: a mechanism for natural user identification
and tracking. In: Proceedings of the international conference on human factors
in computing systems: The Hague, Netherlands; 1e6 April 2000. p. 275e76.
[5] Addlesee M, Jones A, Livesey F, Samaria F. The ORL active floor. IEEE Pers
Comm 1997;4:35e41.
[6] Kwon O, Bahn H, Koh K. A context-aware elevator scheduling system for
smart apartment buildings. In: Proceedings of the 1st international conference
on advances in hybrid information technology: Jeju Island, Korea; 9e11
November 2006. p. 362e72.
[7] Brand ME, Nikovski DN. Risk-averse group elevator scheduling. In: Pro-
ceedings of the 14th international congress on vertical transportation tech-
nologies: Istanbul, Turkey; 27e29 April 2004. p. 1e8.
[8] Lee S, Bahn H. An energy-aware elevator group control system. In: Pro-
ceedings of the 3rd IEEE International conference on industrial informatics:
Perth, Australia; 10e12 August 2005. p. 639e43.
[9] Brand ME, Nikovski DN. Optimal parking in group elevator control. In: Pro-
ceedings of IEEE international conference on robotics & automation: New
Orleans, LA; 26 Aprile1 May 2004. p. 1002e008.
[10] Kim JH, Moon BR. Adaptive elevator group control with camera. IEEE Trans
Ind Electron 2001;48:377e82.
[11] Eguchi T, Hirasawa K, Hu J, Markon S. Elevator group supervisory control sys-
tems using genetic network programming. Congr Evol Comp 2004;2:1661e7.
[12] Igarashi K, Take S, Ishikawa T. Supervisory control for elevator group with
fuzzy expert system. In: Proceedings of the IEEE international conference
industrial technology: Guangzhou, China; 5e9 December 1994. p. 133e37.
[13] Dewen Z, Li J, Yuwen Z, Guanghui S, Kai H. Modern elevator group supervisory
control systems and neural networks technique. In: Proceedings of IEEE in-
ternational conference on intelligent processing systems: Beijing, China; 28e
31 October 1997. p. 528e32.
[14] Pepyne D, Cassandras C. Optimal dispatching control for elevator systems
during uppeak traffic. IEEE Trans Control Syst Technol 1997;5:629e43.
[15] Fujino A, Tobita T, Segawa K, Yoneda K, Togawa A. An elevator group control
system with floor attribute control method and system optimization using
genetic algorithms. IEEE Trans Ind Electron 1997;44:546e52.
[16] Pepyne D, Cassandras C. Design and implementation of an adaptive dis-
patching controller for elevator systems during uppeak traffic. IEEE Trans
Control Syst Technol 1998;6:635e50.
[17] Chung DW, Ryu HM, Lee YM, Kang LW, Sul SK, Kang SJ, et al. Drive systems for
high-speed gearless elevators. IEEE Ind Appl Mag 2001;7:52e6.
[18] Amano M, Yamazaki M, Ikejima H. The latest elevator group supervisory
control system. Mitsubishi Elect ADVANCE 1996;67:88e95.
[19] Maeda T, Komaya K. Revised design calculations of elevator systems. In:
Proceedings of IEEE international conference on systems, man, and cyber-
netics, Nashville: Tennessee, USA; 8e11 October 2000. p. 1041e046.
[20] Chan W, So A, Lam K. Dynamic zoning for intelligent supervisory control. Int J
Elev Eng 1996;1:47e59.
[21] Tobita T, Fujino A, Segawa K, Yoneda K, Ichikawa Y. A parameter tuning
method using genetic algorithms for an elevator group control system. In:
Proceedings of the 22nd international conference on industrial electronics,
control, and instrumentation. Taipei; 5e10 August 1996. p. 823e28
[22] Kaneko M, Ishikawa T, Sogawa Y. Supervisory control for elevator group by
using fuzzy expert system. In: Proceedings of the 23rd international confer-
ence on industrial electronics, control, and instrumentation: New Orleans, LA;
9e14 November 1997. p. 370e76.
[23] Gudwin R, Gomide F, Netto M.A. A fuzzy elevator group controller with linear
context adaptation. In: Proceedings of the IEEE international conference on
fuzzy systems: Anchorage, AK; 4e9 May 1998. p. 481e86.
[24] Kurosawa K, Hirasawa K. Intelligent and supervisory control for elevator
group. Trans Inform Process Soc Jpn 1985;26:278e87.
[25] Qun Z, Zhaoyang W, Youjun Y, Xiaofeng L. The design of multiobjective fuzzy
control algorithm in elevator group control system. In: Proceedings of the 3rd
world congress on intelligent control and automation: Hefei, China; 28 Junee
02 July 2000. p. 2009e012.
[26] Ishikawa T, Miyauchi A, Kaneko M. Supervisory control for elevator group by
using fuzzy expert system which also addresses traveling time. In: Pro-
ceedings of IEEE international conference on industrial technology: Goa, In-
dia; 19e22 January 2000. p. 87e94.
[27] So A, Beebe J, Chan W, Liu S. Elevator traffic pattern recognition by artificial
neural network. Elev Technol 1995;6:122e31.
[28] Markon S, Kita H, Nishikawa Y. Adaptive optimal elevator group control by
use of neural network. Trans Inst Syst Control Inform Eng 1994;7:487e97.
[29] Kim CB, Seong KA, Lee KH, Kim JO. Design and implementation of a fuzzy
elevator group control system. IEEE Trans Syst Man Cybern 1998;28:277e87.
[30] Dai D, Zhang J, Xie W, Yin Z, Zhang Y. Elevator group-control policy with
destination registration based on hybrid genetic algorithms. In: Proceedings
342 O. Kwon et al. / Building and Environment 72 (2014) 332e342
of the IEEE international conference on computer application and system
modeling: Taiyuan, China; 22e24 October 2010. p. 535e38.
[31] Shimbun NS. Elevators get smart. RFID Jpn 2004;1:1e2.
[32] Want R, Hopper A, Falcão V, Gibbons J. The active badge location system. ACM
Trans Inform Syst 1992;10:91e102.
[33] Steinhage A, Lauterbach C. Monitoring movement behavior by means of a
large area proximity sensor array in the floor. In: Proceedings of the workshop
on behavior monitoring and interpretation 2008. pp. 15e27.
[34] Kulkarni AB. Energy consumption analysis for geared elevator modernization:
upgrade from DC Ward Leonard system to AC vector controlled drive. In:
Proceedings of the IEEE industry applications conference: Rome, Italy; 8e12
October 2000. p. 2066e070.
[35] Fujino A, Tobita T, Yoneda K. An on-line tuning method for multi-objective
control of elevator group. In: Proceedings of IEEE industrial electronics, con-
trol, instrumentation, and automation: San Diego, CA; 9e13 November 1992.
p. 795e800.
[36] Doherty L, Simon J, Watteyne T. Wireless sensor network challenges and
solutions. Microwave J 2012;55:22e34.
[37] Christakos C. Sensor networks applied to the problem of building evacuation:
an evaluation in simulation. In: Proceedings of the 15th IST mobile and
wireless summit: Mykonos, Greece; June 2006. p. 134e38.
[38] Díaz-Ramírez A, Tafoya LA, Atempa JA, Mejía-Alvarez P. Wireless sensor
networks and fusion information methods for forest fire detection. Proced
Technol 2012;3:69e79.
[39] Robertson D, Bretherton R. Optimizing networks of traffic signals in real time
e the scoot method. IEEE Trans Veh Technol 1991;40:11e5.
[40] Strang T, Bauer C. Context-aware elevator scheduling. In: Proceedings of the
international conference on advanced information networking and applica-
tions workshops: Niagara Falls, Ont.; 21e23 May 2007. p. 276e81.
[41] Brunette W, Lester J, Rea A, Borriello G. Some sensor network elements for
ubiquitous computing. In: Proceedings of the IEEE/ACM international con-
ference on information processing in sensor networks: Los Angeles, CA; 25e
27 April 2005. p. 388e92.
[42] Zhao H, Liu X. An improved DNA computing method for elevator scheduling
problem. Lect Notes Comp Sci 2013;7719:869e75.
[43] Zhang J, Zong Q. Energy-saving scheduling optimization under up-peak traffic
for group elevator system in building. Energy Build 2013;66:495e504.
[44] Hiller B, Klug T, Tuchscherer A. An exact reoptimization algorithm for the
scheduling of elevator groups. Flex Serv Manuf J 2013;26:1e23.
[45] Gao Y, Lin Y, Sun Y. A wireless sensor network based on the novel concept of an I-
matrix to achieve high-precision lighting control. Build Environ 2013;70:223e31.
[46] Eliades D, Michaelides M, Panayiotou C, Polycarpou M. Security-oriented
sensor placement in intelligent buildings. Build Environ 2013;63:114e21.
[47] Sreedharan P, Sohn M, Nazaroff W, Gadgil A. Towards improved character-
ization of high-risk releases using heterogeneous indoor sensor systems. Build
Environ 2011;46:438e47.
[48] Chen Y, Wen J. Comparison of sensor systems designed using multizone,
zonal, and CFD data for protection of indoor environments. Build Environ
2010;45:1061e71.
[49] Liu X, Zhai Z. Protecting a whole building from critical indoor contamination
with optimal sensor network design and source identification methods. Build
Environ 2009;44:2276e83.