Available online at www.sciencedirect.

com

ScienceDirect
Procedia Computer Science 61 (2015) 507 514

Complex Adaptive Systems, Publication 5

Cihan H. Dagli, Editor in Chief
Conference Organized by Missouri University of Science and Technology
2015-San Jose, CA

Real-Time Optimal Scheduling of a Group of Elevators in a Multi-

Story Robotic Fully-Automated Parking Structure
Jayanta K. Debnath and Gursel Serpen*
Electrical Engineering and Computer Science Department, University of Toledo, Toledo, OH 43606, USA

Abstract

This study presents a simulation-based feasibility study for development of a real-time scheduling algorithm for a multi-story and
fully-automated parking structure with a group of elevators. Each elevator is conceived to carry one vehicle (car, small truck,
SUV or minivan) between floors. Elevator count for a specific parking structure with number of floors in the range of 4 to 20,
and 400 parking spaces on each floor is derived under an assumed customer arrival rate and mean service rate using the waiting
line model of the queuing theory. A scheduling algorithm based on nested partitions and genetic algorithm is evaluated through
the simulation study. The simulation environment models the mean arrival time of customers and elevator dynamics during
morning rush hours for busy urban commercial multi-storied parking structures. Performance evaluation of the implemented
elevator scheduling system was realized using the MATLAB environment. Performance metrics of mean customer waiting and
elevator service times, and maximum customer waiting time were monitored. Simulation results demonstrate that the proposed
design facilitates acceptable customer waiting and service times with good utilization rates for the elevators.

2015
2015TheTheAuthors. Published
Authors. by Elsevier
Published B.V.B.V.
by Elsevier This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of scientific committee of Missouri University of Science and Technology.
Peer-review under responsibility of scientific committee of Missouri University of Science and Technology
Keywords: fully-automated robotic parking, elevator scheduling, real-time, optimal, genetic algorithm, queuing theory, nested partition

1. Introduction

Highly populous metro cities around the globe lack real estate space for many purposes among which parking is
the most prominent. Multi-story parking structures are a promising venue for exploration to address the acute need
for parking spaces. Efficient use of space in a parking structure further requires driving lanes to be eliminated so

* Corresponding author. Tel.: +1-419-530-8158; fax: +1-419-530-8146.

E-mail address: gursel dot serpen at utoledo dot edu

1877-0509 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of scientific committee of Missouri University of Science and Technology
doi:10.1016/j.procs.2015.09.202
508 Jayanta K. Debnath and Gursel Serpen / Procedia Computer Science 61 (2015) 507 514

that they can be used as parking spaces as well. A fully-automated robotic parking structure then becomes a de facto
option to pursue. Transporting vehicles in elevators among the floors is one engineering challenge that needs to be
addressed to make such parking structures a reality.
Extensive research has been conducted for conventional passenger elevator systems in residential structures
particularly for scheduling algorithms. Many heuristic-based scheduling algorithms have been developed [11]-[14].
These methods perform well for specific traffic patterns while being computationally efficient. Another approach is
zoning where each elevator is assigned a number of floors grouped together as a zone [15-17]. The zoning
approach makes it possible to deal with a variety of traffic patterns while also being robust in heavy traffic. Context-
aware elevator scheduling takes advantage of ubiquitous computing and sensor technologies to dynamically chose
and adjust scheduling algorithms based on current and near-future predicted passenger traffic scenarios [18,19].
Optimization of the scheduling process was attempted with genetic algorithms [20-24]. Queue models were applied
in elevator scheduling problems and dynamic programming was used to derive the optimal policy [25]-[26]. In a
destination entry system, passengers can enter their destinations through keyboards before they get into the elevator
cars [10]. For these systems, passenger arrival times, origins, and destinations are known before the systems make
scheduling decisions. For a destination entry system, an exact optimization algorithm was developed in [27] while
dynamic programming and so-called hybrid nested partition and genetic algorithm based methods were developed in
[5,28].
The research reported in this paper considers the elevator scheduling problem in a multi-story, and fully-
automated robotic parking structure [4]. The elevators in such a structure are part of a destination entry system.
Each elevator of this fully automated multi-story parking structure is conceived to carry one vehicle (car, small
truck, SUV or minivan) between floors. This restriction on the carrying capacity of elevators makes it more
challenging to minimize the customer waiting times. We use ideas from the queuing theory, linear programming and
Genetic algorithms to first formulate bounds on the minimum number of elevators and develop an optimal schedule
for elevators during the morning rush-hour period.

2. Methodology

2.1. Elevator scheduling as an optimization problem

The problem domain for the scheduling algorithm is a multi-story parking structure. The parking process is
completely automated. For parking a vehicle, customers drive their vehicles into the parking structure at the entry
or ground floor and leave them at the elevator loading bays. For retrieving a parked vehicle, customers enter the
structure at the ground or entry floor and request their vehicles to be delivered to them at the unloading bays. For
parking, vehicles are loaded onto robotic carts and transported from the entry floor to upper floors using elevators,
where they are moved again by the robotic carts to their designated parking locations. Each elevator can carry only
one robotic cart without or with a vehicle loaded onto it. For retrieval, the requested vehicle, which is already loaded
onto a robotic cart, is brought to the elevator on the same floor and transported by the elevator to the ground or entry
floor. Next, the robotic cart moves the vehicle to the unloading bay for the customer to take delivery and drive
away. We will consider only the parking of vehicles during the morning rush hour within the scope of this paper for
reasons of space.
Consider a parking structure with NF floors and NE elevators. Let NV represents the number of those vehicles,
which are already inside the parking structure awaiting storage. Those vehicles are queued as first-come-first-served:
they are sorted in the ascending order of their arrival times. For each vehicle i, 1i NV, the arrival time tia , the
arrival floor f i a , and the destination floor f i d are tracked at the time of parking or storage requests made by
customers. A two-level integer programming formulation is adapted for elevator scheduling problem in [6], where
each elevator must transport multiple passengers to their destination floors. In our case, the problem is simplified in
that each elevator carries only one vehicle (passenger). Therefore, there is only one assignment consideration,
which is the vehicle-to-elevator assignment. The decision variable is the vehicle-to-elevator assignment, defined as
an NVNE matrix of binary variables, where the (i,j)th element G i, j equals 1 if the i-th vehicle is assigned to the j-th
 Jayanta K. Debnath and Gursel Serpen / Procedia Computer Science 61 (2015) 507 514 509

elevator and 0, otherwise. To guarantee that the decision variable is feasible, the following constraint should be
satisfied: each vehicle must be assigned to one and only one elevator, i.e.
NE
G i , j 1, i, i 1,2,..., NV . (1)
j 1
For vehicles that are already inside the elevators prior to each scheduling iteration, their {Gi,j} values are fixed.
Assignment decisions are only made on those vehicles, which are waiting to be loaded into an elevator.
The objective is to measure the customer satisfaction through the average customer waiting time for delivery or
pickup. Delivery of a vehicle for parking is relatively faster while the wait time for picking up a parked vehicle can
be rather long especially during the rush hours. For a customer who has arrived at the parking structure ground floor
to park his car, the waiting time in the queue until he delivers his car to one of the elevator loading bays forms the
basis for his level of satisfaction. This time period is measured by tid - tia , where tid is the delivery time and tia is
the arrival time. For a customer who is in the queue to pick up a parked vehicle, the waiting time is the time interval
between the arrival time tia and the pickup time tip . This time period, tip - tia , is composed of the retrieval time ( tir )
of the vehicle from its currently parked location to the elevator loading area and the transport time ( tit ) by the
elevator to the ground or entry floor. This ignores two other times: the elevator arrival time at the floor where the
vehicle is currently parked, which may overlap with the retrieval time of the vehicle and therefore may well be zero,
and the waiting time to unload from the elevator and move the vehicle to the pickup bay on the ground floor. In this
study, since we only consider storage requests, the objective function is average customer waiting time for being
able to deliver her vehicle to the queue for the elevator loading bay. The objective function is given by

1 NV d a
J 
ti  ti
NV i 1
 (2)

The overall objective of scheduling is to find a solution for the problem as to minimize the J function in Equation 2
subject to the constraint given in Equation 1. For optimization, we use a hybrid technique based on nested partitions
and Genetic algorithms as discussed next.

2.2. Hybrid Nested Partition and Genetic Algorithm for Elevator Scheduling

We adopt the so-called Hybrid Nested Partitioning and Genetic Algorithm (HNPGA) method to compute a
solution that minimizes the J function in Equation 2 subject to the constraint in Equation 1 [5]. The Nested
Partitioning (NP) method steadily partitions the feasible decision space into subregions, while trying to identify the
most promising subregion. The optimal solution is taught to reside in the most promising subregion, which is
located through sampling. Once the most promising subregion is identified, the NP concentrates on this subregion.
Through iterations, the most promising subregion is gradually reduced by further partitioning and with backtracking.
The promise of the NP method is to compute the optimal solution with probability one, while being a simple and
robust[9]. For the application of NP method, the vehicle-to-elevator scheduling problem can be represented using a
vector of length NV where i-th element of this vector equals j (1jNE) where NE(k) represents the set of available
elevators during the k-th scheduling iteration. Initial feasible space entails all those vectors of length NV where any
j (a specific available elevator) value can be assumed by no more than one i-th element of the vector (representing
the i-th vehicle).
Similar to its application in [5], the Genetic algorithm (GA) is used twice in each iteration of the NP method, one
for selecting the best subregion, and the other for the comparison of the best subregion with the surrounding region.
For both cases, it is used to optimize the assignment of a group of vehicles NV for minimizing objective function
defined in Equation 2. The chromosome is defined as a vector of length NV , where element i (1iNV) equals to
element j (1jNE) if the i-th vehicle is assigned to the j-th elevator. Two types of mutation operators are employed:
(1) random change of the elevator assignment of one vehicle, and (2) random swap of the elevator assignments of
two vehicles. The GA as implemented employs the standard single point crossover operator, which combines two
assignments with good assignment segments for different subgroups of vehicles to generate better assignments. The
510 Jayanta K. Debnath and Gursel Serpen / Procedia Computer Science 61 (2015) 507 514

fitness of each chromosome is defined as the performance of the corresponding assignment per Equation 2. The
pseudocode for the GA is as follows:
x Initialize population with randomly-generated feasible vehicle-to-elevator assignments.
x Expand population through crossover and mutation.
x Evaluate the fitness of individual assignments.
x Select subset of assignments for the next-generation population based on fitness values.
x Repeat previous steps until either a predetermined time period expires or an acceptable quality assignment
is found.

2.3. M/M/S queue model

We employ the M/M/S queue model, which is a system of a single queue (of vehicles in line waiting to be
transported by elevators for parking) with multiple servers (elevators), for our design. According to Kendall [7], it
describes a system where arrivals form a single queue and are governed by a Poisson process; there are NE servers
(elevators) and job service times are exponentially distributed.
An M/M/S queue operation maps to a stochastic process whose state space is the set {0, 1, 2, 3, ...}, where the
value corresponds to the number of customers in the system, including any currently in service. Arrivals occur at a
rate of according to a Poisson process and move the process from one state to its next. Service times have
an exponential distribution with mean service rate of in the M/M/S queue. All NE servers (elevators) serve from
the front of the queue. If there is less than NE jobs (vehicles), some of the servers will be idle. If there are more
than NS jobs, the jobs queue in a buffer. The buffer is of infinite size (which is akin to extending the queue of
waiting vehicles to outside of the parking structure along the side of streets around the city block), so there is no
limit on the number of customers (vehicles) it can contain. The M/M/S queue model can be described as
a continuous time Markov chain and is a type of birthdeath process. Let = /(NE) denote the server utilization
and require the following for the queue to be stable (see reference [8] for a proof):

yields O yields
U  1  o o^O  P u N E ` (3)
P u NE

2.4. Elevator dynamics model

We assume that elevators will reach maximum velocity of VE starting from zero initial velocity with constant
acceleration, aE , after they travel a distance of VE2,max 2a E . Similarly, an elevator needs to travel the same distance
to make a complete stop starting with the maximum velocity and down to zero velocity with constant deceleration of
-aE . There might be two probable scenarios for elevator travel, depending on the travel distance between the starting
and the destination floors.
If the distance between starting and the destination floors is greater than VE2,max 2a E , an elevator travels with
constant acceleration until reaching its maximum velocity permitted by its design. After that, it travels with constant
velocity and starts decreasing its velocity at the point where the distance from the destination is VE2,max 2a E .
Hence, there are three states of motion and they are speeding up, traveling at constant velocity and slowing down as
illustrated in Figure 1 (a). On the other hand, if the distance between the starting and the destination floors is less
than VE2,max 2a E , an elevator goes halfway with constant acceleration and after that (before reaching its maximum
speed) slows down with constant deceleration to a complete stop at the destination floor. There are two travel
modes as speedup, and slowdown as shown in Figure 1 (b).

2.5. M/M/S queuing model and elevator dynamics for scheduling

Customer arrivals are described by a Poisson distribution with a mean arrival rate of (lambda) i.e. average
number of customers arriving per unit of time [2]. This means that the time between successive customer arrivals
 Jayanta K. Debnath and Gursel Serpen / Procedia Computer Science 61 (2015) 507 514 511

follows an exponential distribution with an average of 1/ seconds. The customer service rate is described by a
Poisson distribution with a mean service rate of P (number of customers) i.e. average number of customers that can
be served per unit of time. This means that the service time for one customer follows an exponential distribution
with an average of 1/P. The model assumes that there are NE identical elevators, the service time distribution for
each elevator is exponential, and the mean service time is 1/P seconds.
The total service rate must be greater than the arrival rate, that is PNE> as given by Equation 3. Otherwise, the
waiting line would eventually grow infinitely large. Through this bound, we will formulate the minimum required
number of elevators with respect to number of floors for a given multi-storied parking structure.
The total number of parking spaces on each floor of multi storied parking lot [4] is NCNR. Assuming that each
elevator occupies two spaces, NE elevators will occupy a total of 2NE spaces on each floor. As each elevator has a
loading area, the number of loading areas is NE which will cost an additional NE spaces. The number of available
parking spots at each floor is then calculated as (NCNR)-2NE-NE = NCNR-3NE. Out of these (NCNR-3NE) parking
spots, at least 0.05(NCNR-3NE) parking spots (which is 5% of NCNR-3NE ) must always be open or empty [4]. So
the total capacity of each parking floor is (NCNR-3NE)-0.05(NCNR-3NE)=0.95NCNR-2.85NE parking spots.
Consequently, the capacity (C) of entire multi-storied parking structure in terms of the total number of usable
parking spaces can be defined as NF(0.95NCNR-2.85NE).
Assuming that a morning rush hour period lasts two clock hours for filling the entire parking structure, mean
arrival rate () of vehicles (average number of vehicles arriving per sec interval) is of interest which can be
calculated as the ratio of total number of vehicles arriving per hour to the number of seconds per hour. Total
number of vehicles arriving per hour is equal to the product of average number of vehicles arriving per parking
space per hour (NV,ave) with total number of parking spaces in the parking structure (C). The formula for the
mean arrival rate is then given by = (NV,aveC)/3600. Mean service rate (average number of customers that can be
served per second) of elevators is calculated as
1 1
P , (5)
EMSTR ETT  TL  TU  2TEDO  TEDC 
where EMSTR is the elevator mean service time per request; ETT is the elevator travel time for a distance that is
equal to the halfway height of the parking structure; TL is vehicle loading or embarkation time; TU is vehicle
unloading or dis-embarkation time; TEDO is elevator door opening time, and TEDC is elevator door closing time. We
assume that an elevator needs to travel, on the average, halfway for the multi-storied parking structure to serve a
customer request. The halfway height of a multistoried parking structure is simply the number of floors multiplied
by the height of a floor, where the latter is represented by DF, and is given by DHH=(DFNF)/2.
The time it takes for an elevator to travel a distance of DHH is of interest next. Given the value of DHH and the
maximum velocity of the elevators, VE,max, where the latter is design and technology driven, it is likely that one of
two scenarios as discussed under the Elevator Dynamics will be applicable. Therefore, we will derive the bounds
for both scenarios and use the bound that gives the larger count of elevators for the design and simulation study.
First scenario assumes that, DHH d VE2,max aE holds for which the elevator travel time is given by
ETT=SUT+SDT, where SUT and SDT represent the speed up time and slow down time for the elevator. Let vf,
vi, a, d and t represent final velocity, initial velocity, acceleration, distance traveled, and the time duration of travel,
respectively, then the speed up time can be calculated as follows:

yields
v 2f vi2  2ad  o v f DF N F a E / 2

given that initial velocity is zero: vi 0 ; the acceleration is given by a=aE; and the travel distance is given as
d DHH DF N F / 2 . The travel time can be calculated using

yields
vf vi  at  o t SUT DF N F / 2 a E ,
512 Jayanta K. Debnath and Gursel Serpen / Procedia Computer Science 61 (2015) 507 514

where v f DF N F a E / 2 . Similarly time to slow down can be derived which is same as the time to speed up,
SDT=SUT. Therefore, the total time to move the elevator by the half distance of parking structure height is given by

ETT 2 DF N F / 2aE .

Mean service rate (average number of customers that can be served per second) for this first scenario is given by

1 1 1
P
EMSTR ETT  TL  TU  2TEDO  TEDC   
2 DF N F / 2a E  TEDO  TEDC  TL  TU

 
The second scenario assumes that, DHH ! VE2,max / aE holds for which the elevator travel time is given by
ETT=SUT + CST + SDT, where SUT, CST and SDT represent the speed up time, constant speed time, and slow
down time for the elevator, respectively. Following a similar derivation as the first scenario, the mean service rate
(average number of customers that can be served per second) for the second scenario is given by

VE ,max DF N F
P 1   2TEDO  TEDC   TL  TU
aE 2VE ,max

Now, substituting the values of the mean arrival rate () and the mean service rate (P) in P u N E ! O from the
waiting line model, bounds on the minimum number of elevators for a specific floor count can be derived. It is
further relevant to note that these bounds have been derived by applying the steady-state condition of the M/M/S
queuing model on the fully-automated parking structure.

3. Simulation Study

We have implemented a simulation study for a parking structure with story counts from 4 to 20. Each floor had
the exact same rectangular layout of 2020 with 400 parking spaces. We only considered the morning rush hour
when almost all of the customer requests are for parking their vehicles. Poisson distributed customer arrivals were
simulated according to the arrival scenario in [3]. The simulation was implemented in MATLAB, version 7.11.0
(R2010b), on a desktop PC with the following specifications: the processor is Intel(R) Core(TM) i7 CPU 950 @
3.07 GHz, main memory has 20 GB RAM, and the operating system is Windows 7 Ultimate (64-bit).
Parameters and values used in the simulation study are as follows: DF= 3.5 m, aE=0.7 m/s2, VE,max = 2.5 m/s,
TEDO  TEDC  TL  TU 10 seconds, NV,ave= 0.6175 vehicle per parking space per hour [1], and NC = NR = 20.
Applying values of these parameters on bounds derived in Equation (6) yields the minimum number of elevators for
different floor counts in a parking structure as presented in Table 1. The genetic algorithm (GA) employed the
following values for its parameters [5]: number of generations: Ng=10; population size: Np= 6; crossover probability
= 0.6; and mutation probability = 0.7. Additionally, the nested partition parameters N and Nt are related and defined
as follows to facilitate the assignment of the next 3 vehicles based on the optimization of next 6 vehicles: N=3, and
Nt=6.
We define several performance metrics that will appear in the following figures as follows:
x Waiting Time: The time between a customer vehicles arrival and its pickup by an elevator
x Service Time = Elevator travel time between starting floor and destination floor + 2 (elevator door open
time + elevator door close time) + Vehicle load time + Vehicle unload time
x Scheduling Time: the time required for the HNPGA to schedule NV(k) vehicles to NE(k) available elevators
at iteration k: NV represents only those vehicles which are already in the queue inside the parking
structure awaiting storage.
 Jayanta K. Debnath and Gursel Serpen / Procedia Computer Science 61 (2015) 507 514 513

The most important performance metric is the waiting time for customers. Table 1 presents customer waiting and
elevator service times for floors counts of 4 to 20 along with the required minimum number of elevators. For the
range of floor counts simulated, the worst-case customer waiting time occurs for the 20-story parking structure with
an average of approximately half a minute and a maximum value of 4.5 minutes. The scheduling times presented in
Table * suggests that the system is feasible for real-time operation and there is room to even further speed up the
computations needed for scheduling since MATLAB implementations are comparatively slower than those in
compiled executables. In general, waiting time values are likely to be considered very reasonable by most
customers.

Table 1. Customer waiting and elevator scheduling times.

Number of Floors 4 6 8 10 12 14 16 18 20
Minimum Number of Elevators 17 25 32 39 45 50 56 60 65
Average Waiting Time (sec) 10.8 11.9 14.7 14.1 15.8 19 20.8 21.6 28.8
Maximum Waiting Time (sec) 65.9 68.4 111.5 49.4 50.0 131.4 204.3 143.1 269.0
Average Scheduling Time (sec) 0.1 0.1 0.2 0.2 0.2 0.4 0.4 0.4 0.9
Maximum Scheduling Time (sec) 1.5 2.1 6.1 1.7 5.078 15.7 12.8 14.4 22.8

For a more detailed look at the individual customer waiting and elevator service times for a specific story and
elevator count, Figure 1 is presented. Customer waiting times are mostly small with less than 15 seconds for 4
floors. Waiting times for most customers increase to 20 seconds, 30 seconds and 35 seconds for floor counts of 10,
14 and 20, respectively. Elevator service times are in the neighborhood of 50, 60, 65, and 75 for the same floor
counts. In general, other than a few spikes in customer waiting times, most customers experience a consistently
small wait in all cases. The increase in the elevator scheduling times as the floor count increases appears to be
minimal and does not preclude real-time operation.

 









 









Figure 1.. Customer waiting and elevator service times for various floor and elevator counts

4. Conclusions

This paper presented a simulation study for an elevator scheduling system for a fully-automated robotic and multi-
514 Jayanta K. Debnath and Gursel Serpen / Procedia Computer Science 61 (2015) 507 514

storey parking structure which is intended for large metro areas where real estate is at premium. The presented
study considered the morning rush hour period with nearly all customer requesting to park their cars. Queuing
theory was employed to derive bounds on the minimum number of elevators for a given floor count in the parking
structure. Schedule optimization was accomplished using the nested partitions and genetic algorithms. Simulation
was conducted for floor counts of 4 to 20. Simulation results suggest that the proposed elevator counts will facilitate
real time scheduling with reasonable customer wait times.

References

1. Parking Generation.4th ed. Institute of transport engineer; 2010; p. 201-206.

2. Heizer J, Render B.Operations Management. 9th ed. Pearson Prentice Hall; 2008.
3. Hall RW. Queuing Methods for services and manufacturing. Randolph Hall; 2013.p. 98-104.
4. Serpen G, Dou C. Automated robotic parking systems: real-time, concurrent and multi-robot path planning in dynamic environments.
Applied Intelligence 2015; 42:231251.
5. Sun J, Zhao Q, Luh PB. Optimization of Group Elevator Scheduling With Advance Information. IEEE Transactions on Automation Science
and Engineering 2010; 7(2):352-363.
6. Boulter BT. Elevator Modeling and DC Drive Speed Controller Design. Applied Industrial Control Solutions ApICS LLC; 2010.
7. Kendall DG. Stochastic Processes Occurring in the Theory of Queues and their Analysis by the Method of the Imbedded Markov
Chain. The Annals of Mathematical Statistics 1953; 24(3): 338.
8. Bhat UN. An Introduction to Queueing Theory: Modeling and Analysis in Applications. Birkhauser; 2008.p. 43-50.
9. Shi L, Olafsson S. Nested partitions method for global optimization. Operation Research 2000; 48(3) :390407.
10. Gale J. Destination control and tower top access in Belgium. Elevator World 2002; 10:4549.
11. Bao G, Cassandras CG, Djaferis TE, Gandhi AD, Looze DP. Elevators Dispatchers for Down-Peak Traffic. Technical Report. ECE Dept,
Univ.Massachusetts, Amherst; 1994.
12. Strakosch GR. Vertical Transportation: Elevators and Escalators. New York: Wiley; 1998.
13. Barne GC. Elevator Traffic Handbook: Theory and Practice. New York: Spon; 2003.
14. Hamdi M, Mulvaney DJ. Prioritised A* search in real-time elevator dispatching. Control Engineering Practice 2007; 15(2): 219-230.
15. Chan WL, So ATP, Lam KC. Dynamic zoning in elevator traffic control. Elevator World 1997;3: 136-140.
16. So ATP, Yu JKL, Chan WL. Dynamic zoning based supervisory control for elevators. Proceedings of the 1999 IEEE International
Conference on Control Applications 1999; 2:1591-1596.
17. Strakosch GR. Vertical Transportation: Elevators and Escalators. New York: Wiley; 1998.
18. Strang T, Bauer C. Context-aware elevator scheduling. 21st International Conference on Advanced Information Networking and
Applications Workshops, IEEE 2007; 2:276-281.
19. Kwon O, Bahn H, Koh K. A context-aware elevator scheduling system for smart apartment buildings. Advances in Hybrid Information
Technology. Springer Berlin Heidelberg; 2007.p. 362-372.
20. Corts P, Larraeta J, Onieva L. Genetic algorithm for controllers in elevator groups: analysis and simulation during lunchpeak traffic.
Applied Soft Computing 2004; 4(2): 159-174.
21. Bolat B, Corts P, Yalin E, Aliverii M. Optimal car dispatching for elevator groups using genetic algorithms. Intelligent Automation &
Soft Computing 2010; 16(1): 89-99.
22. 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 Transactions ON Industrial Electronics 1997; 44(4): 546-552.
23. Cortes P, Larraeta J, Onieva L. A genetic algorithm for controlling elevator group systems. Artificial Neural Net, Problem Solving
Methods. Springer-Verlag; 2003.p. 313-320.
24. Tobita T, Fujino A, Segawa K, Yoneda K, Ichikawa Y. A parameter tuning meted for an elevator group control system using a genetic
algorithm. Electrical Engineering in Japan 1998; 124 (1):55-64
25. Liu J, Zhang T, Pian JX, Han F, Wang L. A Queue Dispatch Optimizing Method for Elevator Traffic System. Applied Mechanics and
Materials 2012; 182: 1919-1923.
26. Pepyne DL, Cassandras CG. Optimal dispatching control for elevator systems during up-peak traffic. IEEE Transaction on Control System
Technology 1997; 5(6):629643.
27. Hiller B, Klug T, Tuchscherer A. An exact re-optimization algorithm for the scheduling of elevator groups. Flexible Services and
Manufacturing Journal 2014; 26(4):585-608.
28. Luh PB, Xiong B, Chang SC. Group elevator scheduling with advance information for normal and emergency modes. IEEE Transactions
on Automation Science and Engineering 2008; 5(2): 245-258.