See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/262318594

Discrete-event simulation optimization: A review of past approaches and

propositions for future direction

Conference Paper · July 2013

CITATIONS READS

4 607

1 author:

Linda Riley
University of New Haven
23 PUBLICATIONS   171 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Engineering Education, Assessment; Curriculum and Faculty Development View project

All content following this page was uploaded by Linda Riley on 14 March 2016.

The user has requested enhancement of the downloaded file.

Proceedings SCSC '13 Proceedings of the 2013 Summer Computer Simulation
Conference, Article No. 47 Society for Modeling & Simulation International Vista, CA ©2013
ISBN: 978-1-62748-276-9

Discrete-Event Simulation Optimization: A Review of Past Approaches

and Propositions for Future Direction

Linda Ann Riley, Ph.D.

Roger Williams University
School of Engineering, Computing and Construction Management
[email protected]

Keywords: discrete-event simulation, optimiza- 1. INTRODUCTION

tion, algorithmic optimization techniques
Academicians and practitioners have a num-
Abstract ber of tools to design, measure, study, analyze
and improve large, complex systems. One such
Over the past twenty years, a significant body tool that has been utilized with especially good
of work has been undertaken on the topic of results is simulation. Discrete-event simulation
methods and approaches to optimizing discrete- specifically, allows for the realistic modeling of
event simulation models. Then, as is now, one of stochastic events and the many process varia-
the greatest challenges in optimizing discrete- tions found in most complex, systems. One of the
event simulations is the inability to precisely iden- primary outcomes of discrete-event modeling is
tify “the” optimal solution to a given system mod- improvement in a system’s measures of perfor-
el. This is especially the case as the feasible so- mance
lution space expands. Discrete-event simulation is a widely used tool
Also over the past twenty years, computa- across many disciplines. Although each disci-
tional speed has increased, computing and mod- pline has system specific applications, the goal of
eling costs have decreased and theoretical de- this technique usually involves system analysis
velopments in the field of simulation optimization and/or performance improvement. By simulating
have emerged. Yet a divide appears to be widen- the dynamic nature of a system, one can better
ing. Recent literature indicates a lack of new, in- understand and control random process varia-
novative approaches to optimizing large scale tions. Furthermore, a good simulation that realis-
discrete-event simulation models as well as an tically captures the system under study serves as
absence in addressing the growing chasm be- a model for experimentation. Since almost all
tween the simulation modeling, optimization and large-scale systems are both dynamic and sto-
outcome improvement processes. Many of the chastic in nature, discrete-event simulation is an
studies and advances undertaken in the early to excellent technique to study and analyze these
mid-90’s are those still cited today when discuss- systems.
ing simulation optimization. Algorithmic optimization approaches have
This paper discusses and provides an over- evolved over time as discrete-event simulation
view of theoretical and methodological directions has become more commonplace. Optimization
in discrete-event simulation optimization. In addi- techniques involve numerous dynamic evalua-
tion, it suggests areas of study for advancing the tions of a simulation’s multi-dimensional solution-
field. It is proposed that advances should move space in the search for an optimal solution. Work
the field of study and application in the direction in the area of discrete-event simulation optimiza-
of blurring the boundaries between simulation tion has concentrated for the most part on the de-
modeling, optimization and change implementa- sign and evaluation of various algorithmic and
tion communities instead of widening the gaps. heuristic approaches in searching a simulation’s
solution space. At a higher level of abstraction,
there has been less of a focus on defining optimi-
zation frameworks. Although far outweighed by requiring that input factors be expressed explicit-
the work in algorithmic development, work in the ly; 2) deal well with conditions of local optima
area of optimization frameworks has been under- compared to response surface methods; 3) re-
taken by [Joshi et al. 1996] [Abkay 1996], IEEE, duce computational complexity in contrast to oth-
as well as the Department of Defense. er search techniques, thus reducing solution iden-
tification speed, and; 4) perform quite well under
test conditions comparing a generated optimum
2. REVIEW OF DISCRETE-EVENT with complete enumeration of the solution space.
SIMULATION OPTIMIZATION APPROACHES Other less applied algorithmic approaches for
optimizing discrete-event simulation models in-
Because computational time and cost are crit- clude particle swarm optimization [Clerc 2006]
ical determinants of value and turn-around of the [Olsson 2011], honey bee algorithms [Nakrani
simulation optimization process, a great deal of and Tovey 2006] and fire fly algorithms [Yang
ongoing work over the years has focused on ap- 2009].
plying the most appropriate algorithmic approach Table 1, Overview of Commonly Used Dis-
considering the problem under study. Widely- crete-Event Simulation Optimization Approaches
used search procedures for optimizing a simula- and Algorithms presents the traditionally used
tion’s feasible solution space include: determinis- simulation optimization algorithms with brief
tic search methods; probabilistic search methods comments on the advantages, disadvantages and
and hybrid techniques. processes involved in undertaking each. The ta-
Historically, a great deal of the literature in ble also includes references to a sampling of the
discrete-event simulation optimization is based on seminal work in the area.
the probabilistic search techniques of: simulated
annealing [Liu 1999] [Zolfaghari and Liang 1998] 3. CHALLENGES WITH ADVANCING THE
[Bailey et al. 1997] [Haddock and Mittenthal 1992] KNOWLEDGE BASE OF DISCRETE-EVENT
and evolutionary algorithms [Azadivar et al. 1999] SIMULATION OPTIMIZATION
[Hopper and Turton 1999] [Pierreval and Tautou
1997]. A second area of algorithmic optimization One of the primary drawbacks of the system
development has been in hybrid techniques. This modeling process is the lack of integration be-
approach combines multiple algorithms into a tween the simulation model, the optimization pro-
single optimization strategy [Shi et al. 1999] cess and actions to enact system change as a
[Feyzbakhsh and Matsui 1999] [Gong et al. result of the optimization process. Ultimately the
1997]. goal of modeling many large-scale systems is to
More recently, optimization strategies devel- increase the efficiency with which the system op-
oped using evolutionary or nature-inspired algo- erates as measured by the maximization or mini-
rithms are referred to as metaheuristics [Glover mization of selected parameters of the objective
1986] [Fu et al. 2005] [Glover and Kochenberger function. Looking to the future, more emphasis
2003]. Metaheuristics provide a framework that should be placed on blurring the boundaries be-
overcomes the need to customize an optimization tween the simulation model, optimization and
algorithm for different simulation problems. A change processes. To accomplish this goal, two
number of authors have discussed and explored propositions are advanced in this paper.
the theoretical underpinnings of metaheuristics as
well as various applications [Olafsson 2006] 1. Move from viewing discrete-event simulation
[Yang 2010] [Vasant 2012]. optimization as a static tool to one that is dynami-
The primary reasons why metaheuristic algo- cally integrated into operating practices.
rithms are particularly appropriate for discrete-
event simulation optimization are that these For the most part, the present use of simula-
methods: 1) handle both continuous and discrete tion optimization in large-scale system simulation
input parameters in contrast to search methods scenarios is geared to problems that seek an op-
 Table 1. Overview of Commonly Used Discrete-Event Simulation Optimization
Approaches and Algorithms

Method General Advantages General Disadvantages Process

Intuitive Methods  When used by an indi-  Computational time.   The user selects input parame-
vidual familiar with the Simulation time. ters and undertakes an iterative
system (expert), the  No guarantee or confi- process that involves: 1) varying
method can yield good dence that the ending solu- the parameter levels; 2) complet-
results. tion is the optimal solution. ing a statistically valid number of
 This method is a good  Continuous variables are simulation replications and runs,
one to demonstrate the problematic. and; 3) altering the input parame-
concept of simulation  Difficulty in selecting ters and reevaluating the results.
optimization in a teaching both starting and stopping The objective of this method is to
environment. points for the search. find increasingly better solutions.

Complete Enumer-  Will produce the opti-  Computational time and  Complete factorial experiment
ation mal solution with small cost. of the model is undertaken. Anal-
models defined by a fi-  Works only with discrete ysis of all treatment combinations.
nite solution space. variables.
 Wasted effort due to
testing every feasible solu-
tion in the feasible solution
space.
Tabu Search (see  Deals well with solu-  Not well-developed as a  Feasible solution space is ex-
[Lopez-Garcia et al. tion spaces character- simulation optimization plored by moving from one candi-
1999] [Glover 1977] ized by local optima. methodology. date to its best neighbor. Move-
 Few studies comparing ment occurs even if degradation in
accuracy and precision of the objective function is a result.
results. Tested solutions are considered
 Works only for discrete “tabu” for a user defined number
optimization models. of iterations. Intensification and
diversifications strategies are
used to refine the search direc-
tion.

Pattern Search  Successful search  Does not deal well with  Search moves in direction of
(see [Findler 1987] pattern transferrable to nonunimodality. increasing improvement of the
similar simulation mod- objective function by “steps.” Step
els. sizes vary depending on the sen-
sitivity to change in the objective
function until a user-defined con-
vergence test or tolerance is satis-
fied.
 Method General Advantages General Disadvantages Process

Genetic Algorithms  Relatively fast com-  Genetic algorithms can  Based on the concept of evolu-
[ Hopper 1999] [Col- pared to other search be hard to analyze and tion, genetic algorithms contain
lins 1998] [Aytug et techniques. design depending on the three operators: selection, crosso-
al.1998] [Salzman  Interface process with complexity of the manufac- ver and mutation. The search pro-
and Breitenecker simulation models is turing system being simu- cess involves coding the parame-
1995] [Wellman and easy due to the design of lated. ter set and searching a population
Gemmill 1995] the algorithms.  Recognition of the need of points by means of probabilistic
[Michalewiez 1994]  Does a good job at for more theoretical work in transition rules. The search ends
[Goldberg 1994, identifying the global testing the accuracy of pro- when conditions of a termination
1989] [Holland, optimum in models with duced results. rule are met.
1975] multiple local optima.
 Algorithms are exten-
sible.
 Robust method.
 Low computational
complexity.
 Good building block for
hybrid methods.
Simulated Anneal-  Technique is efficient  Process can require a  Search process involves three
ing [ Liu, 1999] at moving from local op- great deal of computation states: current state, neighboring
[Zolfaghari and Liang tima. time to find the optimal state and optimal states. At each
1998] [ Bailey et  Less computational solution. iteration, a change is made in the
al.1997] [Aarts and time required for each  Attention must be paid to current state and evaluated
Korst, 1989] [Kirk- search iteration however the proper selection of a against a neighboring state by
patrick et al. 1983] more computational time seed solution or current means of cost function. Transi-
required overall because state starting point. tional probabilities and a tempera-
more iterations usually ture parameter dictate the likeli-
are needed. hood of moving from one state to
 Low computational another. The search ends when a
complexity. user-defined number of iterations
 Process avoids cy- or a user-defined number of opti-
cling. mal states is achieved.
 Good building block for
hybrid methods.
Hybrid Techniques  Builds on established  Lack of algorithmic vali-  Process is dependent on the
[ Alireza and Matsui successful algorithmic dation. hybrid technique building blocks,
1999] [ Azadivar and procedures.  Usually not extensible. whether evolutionary strategies,
Tompkins 1999]  Expected lower com-  Interface code can be- simulated annealing, deterministic
[Mason et al. 1999] putational complexity. come problematic depend- searches, or other.
[Fleury et al. 1999]  Expected higher accu- ing on the hybrid technique.
[Shi et al. 1999] racy.  Customization can pre-
[Chen and Gen  Highly customizable clude portability for other
1997] [Ahmed et al., for specific scenarios. manufacturing scenarios.
1998] [Emelyanov  Usually designed to
and Iassinovski handle both discrete and
1997] [Gong et continuous input parame-
al.1997] [Dolgui and ters quite well.
Ofitserov1997]
timal solution at time-specific points. Because of One of the greatest drawbacks however of
the size of the models and the time required to these off-the-shelf simulation packages is the lack
run the model, the process is undertaken as a of flexibility in altering the resident optimization
static event in contrast to an integrated dynamic algorithms. Unless customized code is developed
process. external to the simulation package and then inte-
With the exception of some logistics optimiza- grated into the simulation, the user must take
tion applications involving transportation schedul- what the vendor provides. Depending on the de-
ing, in most manufacturing and service settings, sired optimization function and input parameters,
optimization is not implemented as a dynamic tool the vendor resident procedure may be wholly in-
continuously running in the background and ulti- adequate for the situation under study. Further-
mately driving certain operating decisions. To be more, discrete-event package vendors closely
incorporated as a dynamic tool that contributes to guard as proprietary knowledge, the exact code
intelligent system design, work must continue in used to optimize their products. Selection of a
further integrating the simulation modeling, opti- specific procedure should be dependent on
mization and improvement implementation pro- unique characteristics of the optimization problem
cesses. As optimization algorithms become more and not necessarily what is included in the off-
sophisticated, the simulation optimization process the-shelf simulation software. This is perhaps
appears to be moving further away from the one reason why the move toward metaheuristic
modeling process. This is further exacerbated by frameworks has occurred. These general pur-
the three knowledge domains governing the three pose approaches are evaluated as effective and
processes. For the most part, the large-scale efficient over a range of problems.
system simulation modeling process is owned by To address this shortfall, some type of intelli-
the industrial engineers, operation researchers gent interface is suggested. This interface could
and simulationist community. Optimization algo- be designed to choose from among a number of
rithms and frameworks are driven by the comput- algorithmic optimization procedures based on the
er science community and the improvement im- objective function and input parameters under
plementation processes are owned by the effi- evaluation at any particular moment. This implies
ciency/managerial community. The optimization perhaps an additional layer of AI/neural code that
black box is becoming more and more removed could be incorporated into the optimization pro-
from the ultimate modelers and especially users cess. Ultimately, this intelligent interface could
of the simulation’s results. The gap appears to be “learn” to recognize common optimization scenar-
widening between research and theoretical de- ios, select starting and stopping rules, and poten-
velopment in optimization approaches and appli- tially also interface with the system improvement
cations in contrast to narrowing. framework.
As a further extension to the intelligent inter-
2. Need for simulation optimization procedures to face, dynamic algorithmic visualization capabili-
intelligently recognize input parameters. ties could be incorporated into the optimization
procedures. Immersive technologies are used in
When modeling large-scale system problems, many simulation arenas. Incorporating immersive
the use of off-the-shelf simulation packages is visualization into optimization would serve to
many times necessary. Most of the most popular bring a transparency between the modeling and
discrete-event simulation packages have fast optimization processes. This would allow users
learning curves, are graphically realistic, afforda- and decision makers to interactively view, and po-
ble, and produce easy to read, customized analy- tentially redirect the optimization process. In es-
sis reports. In addition, add-on modules that allow sence, this feature would provide the decision
for external code-writing and customization are maker the ability to immerse him or herself into
common features of today’s off-the-shelf simula- the model, thus “directing” both the simulation
tion packages. In most cases also, these pack- and optimization processes.
ages have built-in optimization modules.
4. CONCLUSION An investigation in shop floor scheduling.”
Annals of Operations Research, Vol. 78 pp. 1-
As a tool to design and improve large-scale 28.
systems, discrete-event simulation optimization is
ideally suited for addressing the complexity asso- Azadivar, F. and Tompkins, G. (1999) “Simulation
ciated with systems characterized as discrete- optimization with qualitative variables and
event and stochastic in nature. With the variety structural model changes: A genetic algorithm
and robustness of algorithmic optimization proce- approach.” European Journal of Operational
dures, virtually all types of system problems can Research. Amsterdam, Feb 16, 1999. Vol.
be modeled and optimized. Yet, even though the 113, Issue 1. pp. 169-192.
algorithmic development in optimization has been
well researched, there appears to be widening Bailey, R.N., Graner, K.M. and Hobbs, M.F.
gaps between the modeling, optimization and im- (1997) “Using simulated annealing and genet-
plementation communities. Furthermore, due to ic algorithms to solve staff-scheduling prob-
lack of transparency among the work of these lems.” Asia-Pacific Journal of Operational Re-
three groups, development of integrative frame- search; Singapore, Vol. 14 Issue 2 pp. 27-43.
works has been lacking. For optimization work to
advance in discrete-event modeling, it is pro- Chen, R. and Gen, M. (1997) “Parallel machine
posed that movement toward the design of dy- scheduling problems using memetic algo-
namically integrated simulation/optimization/im- rithms.” Computers & Industrial Engineering;
plementation products be furthered explored. In Vol. 33; pp. 761.
addition, intelligent optimization interfaces are al-
so proposed for off-the-shelf discrete-event simu- Clerc, M. (2006) Particle Swarm Optimization.
lation packages. Advances in the area of dis- Wiley-ISTE.
crete-event simulation optimization should move
in the direction of blurring the boundaries be- Dolgui, A., Ofitserov, D. (1997) “A stochastic
tween simulation modeling, optimization and method for discrete and continuous optimiza-
change implementation instead of widening the tion in manufacturing systems.” Journal of In-
gaps. telligent Manufacturing, Vol. 8 Issue 5; pp.
405-413.
REFERENCES
Emelyanov, V., Iassinovski, S.I. (1997) “An AI-
Aarts, E. and Korst, J. (1989) Simulated Anneal- based object-oriented tool for discrete manu-
ing and Boltzmann Machines: A Stochastic facturing systems simulation.” Journal of In-
Approach to Combinatorial Optimization and telligent Manufacturing, Vol. 8 Issue 1; pp.
Neural Computing. John Wiley and Sons, 49-58.
Chichester, U.K.
Feyzbakhsh, A. and Matsui, M. (1999) “Adam-
Ahmed, M., Alkhamis, T. and Miller, D. (1998) Eve-like genetic algorithm: a methodology
“Discrete Search Methods for Optimizing Sto- for optimal design of a simple flexible as-
chastic Systems.” Computers & Industrial sembly system.” Computers & Industrial
Engineering, Vol. 34, pp. 703-716. Engineering, Vol. 36 pp. 233-258.

Abkay, Kunter S. (1996) “Using simulation optimi- Findler, N. V., Lo, C. and Lo, R. (1987) “Pattern
zation to find the best solution.” IIE Solutions; Search for Optimization.” Mathematics and
Norcross, Vol. 28, Issue 5 pp. 24-29. Computers in Simulation 29, no. pp. 41–50.

Aytug, H., Bhattacharyya, S. and Koehler, G.

(1998) “Genetic learning through simulation:
 th
Fleury, G., Goujon, J., Gourgand, M., Lacomme, Presented at the 17 ASEM National Con-
P. (1999) “Multi-agent approach and sto- ference, Dallas.
chastic optimization: random events in
manufacturing systems.” Journal of Intelli- Liu, J. (1999) “The impact of neighborhood size
gent Manufacturing, Vol. 10 Issue 1; pp. 81- on the process of simulated annealing:
101. Computational experiments on the flowshop
scheduling problem.” Computers & Industrial
Glover, F. and Laguna, M. (1997) Tabu Search, Engineering, New York; Vol. 27 Issue 1,2;
Kluwer, Boston. pp. 285-288.
Glover F. (1986) “Future paths for integer pro- Lopez-Garcia, L., and Posada-Bolivar, A. (1999)
gramming and links to artificial intelligence.” “A simulator that uses Tabu Search to ap-
Computers and Operations Research, Vol.13, proach the optimal solution to stochastic in-
pp. 533-549. ventory models.” Computers & Industrial En-
gineering, Vol. 37 Issue 1,2; pp. 215-218.
Glover F. and Kochenberger G. A. (2003) Hand-
book of Metaheuristics, Springer. Mason, A., Ryan, D. and Panton, D. (1999) “Inte-
grated simulation, heuristic and optimisation
Goldberg, D.E. (1989) Genetic algorithms in approaches to staff scheduling.” Operations
search, optimization, and machine learn- Research, Vol. 46 Issue 2; pp. 161-178.
ing. New York, NY: Addison Wesley.
Michalewiez, Z. (1994) “Evolutionary compu-
Goldberg, D. E. (1994) “Genetic and evolution- tation techniques for non-linear program-
ary algorithms come of age.” Association of ming problems.” International Transac-
Computing Machinery Communications of tions in Operational Research. Vol. 1 Is-
the ACM, Vol. 37, Issue 3, pp. 113-122. sue 2; pp. 233-240.
Gong, D., Gen, M., Yamazaki, G. and Xu, W. Nakrani, S and C. Tovey (2004) “On honey bees
(1997) “Hybrid evolutionary method for ca- and dynamic server allocation in Internet
pacitated location-allocation problem.” hosting centers,” Adaptive Behavior, 12, 223-
Computers & Industrial Engineering, Vol. 240.
33, pp. 577-80.
Olafsson, S. (2006) “Metaheuristics,” in Nelson
Haddock, J. and Mittenthal, J. (1992) “Simulation and Henderson (eds.) Handbook on Simula-
optimization using simulated annealing.” tion, Elsevier, 633-654.
Computers & Industrial Engineering. Vol. 22
Issue 4; pp. 387-395. Olsson, A., ed. Particle Swarm Optimization:
Theory, Techniques and Applications, Nova
Holland, J.H. (1975) Adaptation in natural and Science Publishers Inc, 2011.
artificial systems. MIT Press.
Pierreval, H. and Tautou, L., May. (1996) “Using
Hopper, E. and Turton, B. (1999) “A genetic al- evolutionary algorithms and simulation for the
gorithm for a 2D industrial packing problem.” optimization of manufacturing systems.”
Computers & Industrial Engineering, Vol. 37 French Institute of Mechanical Engineering
Issue 1, 2; pp. 375-378. (IFMA) Campus des Cezearx, B.P. 265.
F63175 Aubiere Cedex France.
Joshi, B.D., Unal, R. White, N.H., and Morris,
W.D. (1996) “A framework for the optimiza- Salzman, M. and Breitenecker,. F. (1995) “Ge-
tion of discrete event simulation models.” netic algorithms in discrete event simulation,”
 in Proceedings of the Eurosim. Congress 95. tion of Associate Department Head for the De-
September 11 - 15. Vienna. Austria. pp. 213- partment of Industrial Engineering at New Mexico
218. State University (NMSU). In addition, she served
as the founder and Director of the Advanced
Shi, L., Olafsson, S. and Chen Q. (1999) “A new Modeling and Simulation Laboratory at NMSU
hybrid optimization algorithm,” Computers & and Director of a university-wide economic devel-
Industrial Engineering, Vol. 36; pp. 409-426. opment research center.
Dr. Riley has extensive business and engi-
Shi, L., Olafsson, S. and Sun N. (1999) “New neering consulting experience. As well, she is an
parallel randomized algorithms for the travel- active researcher, teacher and author in the area
ing salesman problem,” Computers & Opera- of simulation modeling and large-scale system
tions Research, Vol. 26; pp. 371-9. optimization. She has taught over 30 different
courses in her career, many of them simulation
Vasant, P. (2012) Meta-Heuristics Optimization focused and has written or co-authored over 120
Algorithms in Engineering, Business, Eco- academic/research publications and over 150 re-
nomics, and Finance. Edited by Pandian search proposals.
Vasant. 1st ed. IGI Global. Dr. Riley earned an M.S. in Industrial Engi-
neering as well as a Ph.D. in Logistics from New
Yang, X. (2010) Engineering Optimization: An Mexico State University, completed a two-year
Introduction with Metaheuristic Applications, post graduate fellowship at Brown University,
1st ed. Wiley. earned an MBA from Suffolk University and an
undergraduate degree from Boston University.
Yang X. S. (2009) “Firefly algorithms for multi-
modal optimization,” 5th Symposium on Sto-
chastic Algorithms, Foundation and Applica-
tions (SAGA 2009) (Eds. Watanabe O. and
Zeugmann T.), LNCS, 5792, pp. 169–178.

Wellman, M.A. and Gemmill, D.D. (1995) “A

genetic algorithm approach to optimization
of asynchronous automatic assembly sys-
tems,” International Journal of Flexible
Manufacturing Systems, Vol. 7 Issue 1;
pp.27

Zolfaghari, S. and Liang, M. (1998) “Machine

Cell/Part Family Formation Considering Pro-
cessing Times and Machine Capacities: A
Simulated Annealing Approach,” Computers
& Industrial Engineering, Vol. 34, No. 4; pp.
813-823.

Biography

Dr. Linda Ann Riley is currently Engineering

Program Coordinator and Professor of Engineer-
ing for the School of Engineering, Computing and
Construction Management at Roger Williams
University (RWU). Previously, she held the posi-

View publication stats