International Journal Of Advanced Engineering Research and Science (IJAERS) [Vol-4, Issue-3, Mar- 2017]

https://dx.doi.org/10.22161/ijaers.4.3.17 ISSN: 2349-6495(P) | 2456-1908(O)

A Review on Simulation Optimization

Mobin Ahmad

Department of Mathematics, Faculty of Science, Jazan University, Saudi Arabia

Abstract One of the primary and most important

(Figure
employments of simulations is for optimization. Simulation
1). Simulation optimization entails finding optimal settings
optimization can be characterized as the way toward
finding the best info variable qualities from among all of the input variables, i.e. values of which
potential outcomes without unequivocally evaluating each optimize the output variable(s).
possibility. The goal of simulation optimization is to
minimize the assets spent while boosting the data acquired
in a simulation experiment. The purpose of this paper is to
review the zone of simulation optimization. A critical review
of the methods employed and applications developed in this
generally new range are introduced and striking victories Fig.1: A Simulation Model
are highlighted. Simulation optimization software tools are
discussed. The target group is simulation practitioners and Such problems emerge habitually in engineering, for
theoreticians and additionally fledglings in the field of example, in process design, in mechanical experimentation,
simulation. in design optimization, and in reliability optimization. This
Keywords Simulation, Optimization, important, process, is the issue we will address in this paper. A simulation
resources, information, methods, develop, successes, optimization model is shown in Figure 2. The yield of a
software tools. simulation model is utilized by an optimization strategy to
give criticism on advancement of the quest for the optimal
I. INTRODUCTION solution. This thus manages further contribution to the
The mathematical model of a system is concentrated on simulation model.
using simulation; it is known as a simulation model. System
behavior at particular estimations of info factors is assessed
by running the simulation model for a settled timeframe. A
simulation experiment can be characterized as a test or a
progression of tests in which significant changes are made
to the information factors of a simulation model so that we Fig.2: A Simulation Optimization Model
may observe and recognize the purposes behind changes in
the output variable(s). At the point when the quantity of II. REVIEW OF LITERATURE
information factors is huge and the simulation model is Simulations Optimization of true occasions can permit a
perplexing, the simulation experiment may turn out to be complex problem to be dismembered and examined in a
computationally restrictive. Other than the high productive, safe, and financially savvy way. A simulation
computational cost, a much higher expense is brought about becomes a much more valuable instrument when optimizing
when imperfect info variable qualities are chosen. The way an arrangement of parameters, especially in circumstance
toward finding the best info variable qualities from among where experiments on this present reality framework are
all potential outcomes without unequivocally evaluating troublesome or impractical. Simulation optimization, as a
each plausibility is simulation optimization. The goal of rule, tries to minimize an objective function:
simulation optimization is minimizing the assets spent
while amplifying the information acquired in a simulation
experiment. Where represents an input vector of parameters,
A general simulation model comprises n input variables is the scalar objective function and is the constraint
set [1, 2]. The info parameters are frequently alluded to as
and m output variables

www.ijaers.com Page | 116

International Journal Of Advanced Engineering Research and Science (IJAERS) [Vol-4, Issue-3, Mar- 2017]
https://dx.doi.org/10.22161/ijaers.4.3.17 ISSN: 2349-6495(P) | 2456-1908(O)
variables, while the yield exhibitions are known as the converge to the minima as the step size gets sufficiently
reactions [3]. Simulations can be subdivided into two smaller [2, 4]. The difficulty in using this approach is
classifications of element variable sorts: consistent or estimating the gradient, which will not be continuous for
discrete. In continuous simulations, the limitations are discrete simulations. The most common gradient
normally the set or every single genuine number, n . approximation method is the method of finite differences,
In discrete event simulations, the response function is not where a small number of output values are taken for small
specifically accessible, may not be nonstop, or may not be changes in the simulation parameters. The two-sided,
in shut structure, and subsequently standard mathematical central difference gradient operator, for example, is:
extremism solutions cannot be utilized. In this manner, a
hunt must be performed over the discrete factor set [1, 2, 3,
and 4]. This comprises of picking a worth for all
parameters, running the simulation until the appropriate Where is the perturbation of input parameter i and i p is
stopping criteria are met, and after that measuring the yield a vector with a one in the ith place and zeros elsewhere.
of the simulation that is to be optimized. In simulations with This is essentially taking the discrete derivative, or slope
numerous parameters, a comprehensive inquiry turns into a of the function for each separate dimensional input
period expending errand. Propels in computational power, parameter, denoted by i [2, 4, 8]. However, if the simulation
consolidated with more current strategies for decreasing the output is noisy, then the gradient estimation could also be
pursuit space, have permitted discrete optimization noisy, possibly making the decision point move in an
techniques to be deployed with more success. inappropriate direction [4, 7].
It is clear that simulations with a higher number of variable
input parameters will require more calculations to appraisal
the slope. In particular, utilizing the focal distinction
inclination, it will take 2q simulation measurements, where
q is the dimensionality, or number of variable input
parameters [2, 8]. Another method has been proposed to
lessen the quantity of simulation measurement required.
This technique is known as Infinitesimal Perturbation
Fig.3: Six Domains of Simulation Optimization Analysis (IPA), and is summarized in [5] by the formula:

Simulation Optimization has also become an integral part of

many commercially available software simulation packages.
Now, the vector x becomes a random perturbation vector,
Bowden and Hall [5] have described six domains which
and the gradient estimation requires only two simulation
should be addressed when designing automatic simulation
measurements, regardless of the input vector
optimization tools (figure 3).
dimensionality. Thus, the speedup using this gradient
estimator is potentially q. However, this potential is realized
III. GRADIENT AND STOCHASTIC METHODS
only if the number of iterations required for convergence is
Stochastic approximation methods (SAM) endeavor to
not increased. It is also been shown that the x vector
discover minima by moving toward the steepest slope of the
should be independently and symmetrically distributed
function. This is an iterative process, where every cycle
about zero with finite inverse moments. This can be
comprises of evaluating the angle of the simulation model at
achieved by using the Bernoulli 1 distribution.
the current decision point, and after that moving this
decision point along the slope with a specific stride size.
IV. SIMULATION OPTIMIZATION
This development of the decision point can be expressed as
APPLICATIONS
Simulation optimization methods have been connected to
applications with a solitary target, applications that require
Where the current decision point solution is is the optimization of multiple criteria, and applications with
the estimate of the gradient, n is the step size, and is a non-parametric objectives.
mapping onto the set [2, 4, 11]. This method has received Azadivar et al. (2010) connected a simulation optimization
much attention, mainly because it has been proven to algorithm based on Box's perplexing hunt strategy to

www.ijaers.com Page | 117

International Journal Of Advanced Engineering Research and Science (IJAERS) [Vol-4, Issue-3, Mar- 2017]
https://dx.doi.org/10.22161/ijaers.4.3.17 ISSN: 2349-6495(P) | 2456-1908(O)
optimize the locations and stock levels of semi-completed learning of capacity or information imperatives [9]. Genetic
items in a force sort production system. Algorithms endeavor to iteratively discover an all inclusive
Corridor et al. (2009) utilized ES with a simulation model optimum solution by investigating the reaction surface of
for streamlining a kabana sizing problem. Dad technique to the simulation, and developing the best solutions in a
stock models where the interest has a related restoration comparable way to Darwin's hypothesis of evolution. A
arrival process. conceivable solution is encoded as a chromosome, with
Tompkins and Azadivar (2011) proposed an approach to every quality in a chromosome speaking to a variety of a
join a GA and an article oriented simulation model solitary input parameter. The wellness of a chromosome
generator to locate the ideal shop floor design. an approach speaks to how close the chromosome's qualities will convey
to consolidate the process plant production operations into the simulation to its optimum value. A chromosome with
the design of an office by joining simulation and GA. a low wellness will have a higher likelihood of being expelled
calculation that joined SA and simulation to locate a from the population. A population is a gathering of
suitable dispatching need of operations to minimize the chromosomes in one algorithm iteration. GA requires two
aggregate lateness for a business flexible manufacturing operations, cross over and mutation, to change over one
system (FMS). population of chromosomes to the following [3, 4]. The
essential stream of hereditary calculations is demonstrated
V. SIMULATION OPTIMIZATION SOFTWARE in figure 4.
A longstanding goal among a portion of the simulation
practitioners and theoreticians was having the capacity to
direct a progression of simulations in the most effective way
as opposed to performing "blind" analyses and accepting
that no less than one of the tests will yield the best
contrasting option to execute (Glover et al., 2012).
Numerous simulation software developers today have
turned out to be more mindful of the importance of finding
optimal and close optimal solutions for applications in
minutes, instead of playing out a comprehensive
examination of pertinent options in days or months.
Simulation software that incorporates extraordinary hunt
techniques to control a progression of simulations to
uncover optimal or close ideal scenarios includes: Pro- Fig.4: Genetic Algorithm flow chart
Model, Auto-Mod, Micro Saint, Lay-OPT, and Factory
OPT. A brief portrayal of every softwares optimization Some of the issues when implementing GAs include gene
and/or statistical module follows. representation, crossover operator selection, and mutation
The extra optimization module for Pro-Model is called Sim operator selection. Binary strings are a very common choice
Runner Optimization. This module consists of two elements for gene representation, because they are very general, can
for investigating and optimizing existing Pro-Model be used for any size data type, and require minimal storage.
simulation models. The primary component is a factorial An example of a binary string chromosome with genes
configuration of tests that uncovers the impact of an representing byte sized parameters X, Y, and Z is given in
adjustment in info element on the objective function. The figure 5.
second feature is a multi-variable optimization that tries
different combinations of input components to land at the
combination that yields the best objective function value.
Fig.5: Chromosome of 3 bytes
VI. GENETIC ALGORITHMS
Genetic Algorithms (GA) are a subset of Evolutionary The crossover operator is in charge of making new
Algorithms, which are main stream in optimization chromosomes from two existing ones. A famous way to
literature on account of their generality. In particular, they deal with this is to choose a little number of bits from every
just require the Monte-Carlo simulation output, with no quality and essentially swap them between the two

www.ijaers.com Page | 118

International Journal Of Advanced Engineering Research and Science (IJAERS) [Vol-4, Issue-3, Mar- 2017]
https://dx.doi.org/10.22161/ijaers.4.3.17 ISSN: 2349-6495(P) | 2456-1908(O)
chromosomes [3, 4]. The mutation operator is basic since it Reducing Factors in Optimization via Discrete-Event
permits the changing of qualities to keep a solution from Simulation, Proc. of the Winter Simulation
being caught in a local optimum. This can be expert by Conference, Monterey, CA, 2008.
arbitrarily selecting a bit from every chromosome, and [7] A. Persson, H. Grimm, A. Ng, T. Lezama, J. Ekberg,
upsetting it with a predefined probability distribution [3, 4]. S. Falk, and P. Stablum, Simulation-Based Multi-
As the iterations of the calculation increase, chromosomes Objective Optimization of Real-World Scheduling
(solutions) with lower wellness will be evacuated in the Problem, Proc of the Winter Simulation Conference,
hybrid stage, so populaces will comprise of more Monterey, CA, 2012.
chromosomes with higher wellness. The calculation can end [8] A. Persson, H. Grimm, and A. Ng, On-Line
when a sought number of cycles has been come to, or the Instrumentation for Simulation-Based Optimization,
standard deviation of a populations fitness has been Proc. of the Winter Simulation Conference, Monterey,
minimized. CA, 2011, 304-311.
[9] Azadivar, F., J. Shu, and M. Ahmad. 2010. Simulation
VII. CONCLUSION Optimization in Strategic Location of Semi-Finished
We have provided an introduction to simulation Products in a Pull-Type Production System,
optimization, with emphasis on gradient-based techniques Proceedings of the Winter Simulation Conference,
for continuous parameter simulation optimization and on 1123-1128.
random search methods for discrete parameter simulation [10] Hall, J. D., and R. O. Bowden. 2011. Simulation
optimization. Although simulation optimization has Optimization for a Manufacturing Problem,
received a fair amount of attention from the research Proceedings of the Southeastern Simulation
community in recent years, the current methods generally Conference Huntsville, AL, Society for Computer
require a considerable amount of technical sophistication on Simulation, 135-140.
the part of the user, and they often require a substantial [11] Tompkins, G. and F. Azadivar. 2011. Genetic
amount of computer time as well. Therefore, additional Algorithms in Optimizing Simulated Systems,
research aimed at increasing the efficiency and ease of Proceedings of the Winter Simulation Conference,
application of simulation optimization techniques would be 757-762.
valuable. [12] Glover, F., J. P. Kelly, and M. Laguna. 2012. New
Advances and Applications of Combining Simulation
REFERENCES and Optimization, Proceedings of the Winter
[1] M.C. Fu, F. W. Glover, and J. April., Simulation Simulation Conference, 144-152.
Optimization: A Review, New Developments, and
Applications, Proc. Of the 2010 Winter Simulation
Conference, Orlando, FL, 83-95.
[2] S. Olafsson, and J, Kim, Simulation Optimization,
Proc. Of the Winter Simulation Conference, San
Diego, CA, 2011, 79-84.
[3] J. April, et al, Practical Introduction to Simulation
Optimization, Proc. of the Winter Simulation
Conference, New Orleans, LA, 2008, 71-78.
[4] T. M. Alkahmis, and M. A. Ahmed, Simulation- Based
Optimization Using Simulated Annealing With
Confidence Interval, Proc of the 2007 Winter
Simulation Conference, 514-519.
[5] A. A. Prudius, and S. Andradottir, Two Simulated
Annealing Algorithms for Noisy Objective Functions,
Proc of the 2010 Winter Simulation Conference,
Orlando, FL, 797-802.
[6] J. A. B. Montevechi, R. G. A. Filho, and A. L.
Medeiros, Application of Factorial Designs for

www.ijaers.com Page | 119