Journal of Cleaner Production 85 (2014) 433e441

Contents lists available at ScienceDirect

Journal of Cleaner Production

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

Increasing transport efficiency using simulation modeling in

a dynamic modeling approach
Girish Upreti*, Prasanna V. Rao, Rapinder S. Sawhney, Isaac Atuahene, Rajive Dhingra
Department of Industrial and Systems Engineering, The University of Tennessee, Knoxville, TN, USA

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

Article history: The effective movement of people is the purpose of transportation, but the behavior of people within a
Received 9 July 2014 transportation system is not always well understood. Using analytical data, particularly simulations, to
Received in revised form understand the behavior of passengers can offer valuable insight on improving processes and optimizing
29 August 2014
transportation infrastructure. The bus route at the University of Tennessee, Knoxville, USA, was opti-
Accepted 1 September 2014
mized and the optimal number of buses on the route was estimated in order to streamline the working
Available online 16 September 2014
process and reduce any unwanted activity, thereby saving operating costs and reducing pollution. The
present paper presents a dynamic simulation modeling methodology that reduces the variation between
Keywords:
Transportation
the simulation model results and actual system performance. The dynamic simulation model presented
Dynamic simulation updates itself based on the dynamics of the actual system to reflect correctly the impact of organization
Bus route restructuring on overall organizational performance.
© 2014 Elsevier Ltd. All rights reserved.

1. Introduction university model whose implementation is a process of continuous

improvement in environmental, social, and economic performance
Transport is one of the strategic policy areas in which cities can that should be made through incremental steps. This requires
influence their effect on climate change. Inefficient transportation major efforts from key members of the university community
systems increase pollution and waste money. By using analytical including additional funds and resources. Lozano also suggests that
data to understand passenger behavior and frequency of trans- small steps can create sustainability in universities. Fuel cells for
portation modes, routes in a transportation system can be adjusted transportation and distributed power can be a better option for
to reduce both pollution and costs. For example, instead of four reducing pollution. The government can continue providing sup-
partially full buses running back-to-back, two buses may be able to port for fuel cells technology so that it can compete with traditional
handle a given route. However, understanding the variety of factors technology as proposed by Greene et al. and Upreti et al.
that lead to an individual's decision to use mass transit, ranging To create an accurate analytic model, a transportation authority
from travel time and weather conditions to the purpose of a trip has to consider all of these factors. Simulations can be used to
and the availability of transportation options, is helpful. analyze passenger behavior and transportation routes; based on
Barnes and Jerman suggested that Environmental Management those simulations, further recommendation for adjusting vehicle
Systems (EMS) can be an important tool for educational institutions frequency and routes can be made. However, two main problems
to effectively manage diverse environmental issues and improve can seriously compromise the usefulness of simulations: the first is
campus sustainability. According to Savely et al., only two EMS faulty data input and the second is the selection of inappropriate
models have been proposed specifically for colleges and univer- input distributions.
sities, the Osnabruck model and the university of South Carolina If data are not input correctly, imperfect simulations will result
sustainability initiative. Velazquez et al. developed a sustainable regardless of how long the simulation run is. The most common
reason for incorrect data input is the lack of actual system knowl-
edge by the simulation analyst. To avoid collecting invalid data, it
has been recommended to involve managers of the actual system in
* Corresponding author. Department of Industrial and Information Engineering,
University of Tennessee, 525K John D. Tickle Building, 851 Neyland Drive, Knoxville,
each step of simulation modeling (Banks, 1999; Robinson 2005;
TN 37996-2315, USA. Tel.: þ1 865 974 3333; fax: þ1 865 974 0588. Robinson et al., 2001; Robinson, Edwards et al., 1998). In addition,
E-mail addresses: [email protected], [email protected] (G. Upreti). it has been found that where random inputs are considered by

http://dx.doi.org/10.1016/j.jclepro.2014.09.002
0959-6526/© 2014 Elsevier Ltd. All rights reserved.
434 G. Upreti et al. / Journal of Cleaner Production 85 (2014) 433e441

simulation analysts and where such stochastic inputs are provided, replica. Failure to adequately implement the conceptual model will
the output will also be random (Kelton, Sadowski et al., 2009; result in erroneous results; thus, a large possibility exists to over-
Kelton, 1997). look the conceptual model, as the results are not validated until
When collected data are improperly analyzed and the wrong later stage, and at that point fixing the model becomes a time-
distribution is considered, the result is an unrealistic system rep- consuming and costly affair.
resentation in a simulation model. Data collection is a challenging process as it is time
The second main problem in simulations is that appropriate consuming and data may not be easily available. If collecting data
input distributions must be selected to characterize the stochastic becomes challenging, then efforts should be made to conduct a
behavior of the modeled system (Wagner and Wilson, 1995). Failure survey. Most of the time, interaction with the people who are
to select appropriate input distributions can result in misleading handling the system will yield helpful results, remembering that
simulation output and thus poor system design decisions (Chick, any source of data must be documented for validation. Once the
2001). data are collected, proper analysis should be conducted in terms
A factor that will affect the accuracy of simulation results over of fitting distributions, schedules, and process times. Surveys and
the long run is the common practice of considering visible short- onsite observations are suitable methods for collecting data.
term constraints and data in systems where the simulation model Critical events are defined as events which arise when oper-
is expected to perform for a long period of time (Banks, Nelson ations happen over a long period of time, and addressing them
et al., 2009). Most real-life operational models are dynamic in na- with the right solution logic is necessary for proper functioning
ture, and their system state variables change frequently, with time of the model. After identifying these events, an appropriate so-
affecting system behavior. When systems of such kinds are simu- lution to these events must also be decided upon and docu-
lated for the long run, the variation is wrongly assumed to be mented. The role of the programmer in this step is developing
negligible, affecting the accuracy of the simulation model (El-Haik the system logic to the critical events in software.
and Al-Aomar 2006). In other words, the changes that happened in Different areas of operations have their own specialty soft-
the actual system or processes over the extended period of time are ware and the choices available today are numerous. So, according
not considered in the simulation model and thus the results are not to the field of the actual system, suitable software has to be
accurate and the reliability of the simulation model is affected selected and the simulation logic has to be developed. Dynamic
(Kelton et al., 2009; Ming et al., 1990). simulation models usually require the ability to read and write to
There are no such models that explicitly enhance the accuracy of an external metadata program such as MS EXCEL. The logic of the
a simulation model over long runs. However, in the present study, simulation model has to be checked by personnel handling the
an intelligent simulation modeling framework that changes its rule system to verify the system details. A variety of inputs is used to
priority over a simulation run is developed, answering the need for generate results, and the outputs are compared to manual cal-
developing a simulation methodology that is generic in nature in culations and actual system performance (Fishman and Kiviat,
order to increase the accuracy of simulation modeling over long 1967).
runs.
Validating simulation models is necessary to determine
whether a model is an accurate replica of an actual system or not.
2. Methodology
Invalid models lead to erroneous data. Various statistical tech-
niques, such as hypothesis testing and sensitivity analysis, can be
A dynamic simulation modeling methodology is a generic
used to validate the results.
framework that can be adapted to any system in order to achieve
For each configuration of interest, the number of replications,
accurate simulation models. The purpose of developing the present
the sample size, the run length, and warm-up periods have to be
methodology is to address the influence of time-based critical
decided. Simulation runs must be performed to check all the
events in the simulation model. This research proposes a dynamic
critical conditions under a variety of inputs. In dynamic simula-
simulation methodology that is generic in nature and can be
tion modeling with a variety of different inputs, the system logic
adapted to any dynamic work environment, obviating the well
will re-route to address each condition. So increasing the number
known limitations of existing simulation methodology. The dy-
of replications over a wide range of input parameters will ensure
namic simulation modeling methodology is developed from
that all possible combinations are tested over time. Statistical
traditional simulation methodologies using the concept of contin-
analysis of the end results shows the effectiveness of the simu-
uous process improvement (Manuj et al., 2009). The steps to
lation model.
develop a dynamic simulation methodology, in which critical
Analyzing the simulation output data depends to a great
events are induced in the simulation over long runs, are as follows:
extent on the type of the simulation model and its objectives.
The first step involves defining the overall objectives and the
Traditionally, simulation output data are analyzed in the form of
reason for developing the simulation model. Defining the problem
confidence intervals, or confidence regions in cases of multivar-
statement helps to maintain the overall focus of the simulation
iate output. Comparing different simulation methods, using vi-
logic and will help avoid the failure of the model during the analysis
sual inspection of graphs and their metrics such as mean, lower
phase.
and upper limits, standard deviation and percentiles, are
The next step is to identify the independent and dependent
preferred methods (Kelton, 1997).
variables. Performance criteria are measured by dependent vari-
ables, and system parameters are represented by independent
variables. Manipulating the independent variables significantly 3. Development of dynamic simulation model for
affects the dependent variables. Independent and dependent var- transportation department
iables also help to identify the critical events and control for them
over time. The generic dynamic simulation steps for developing a dynamic
The conceptual model ensures that the observed criteria fall simulation model can be adapted to any possible working scenario.
well within the problem statement. Conceptual simulation is used Each operation has its own unique requirements, and the dynamic
for system understanding, validation, and teaching. Data are added simulation steps can be modified to suit all conditions. To illustrate
later on to the conceptual model to validate and fine-tune the the versatility of the dynamic simulation modeling, a university
 G. Upreti et al. / Journal of Cleaner Production 85 (2014) 433e441 435

campus transportation route is modeled on a long run, and the 4. Case study
dynamic simulation steps are tested.
Fig. 1 illustrates the methodology for implementing a dynamic The following case study is based on a highly successful opti-
simulation model. The generic dynamic simulation methodology mization and process improvement project that was completed for
has been carefully adapted to be used in the transportation model. the University of Tennessee at Knoxville (UTK) transit system.
The transportation department at University of Tennessee (UT) has Traditional simulation was used to replicate the fast track route
already been modeled in the conventional way to replicate the between the main university campus and the agriculture campus,
system and find the optimal number of buses. In order to consider which is also known as the Ag Campus. The purpose of the study
long-run implications for the same system, it is necessary to was to demonstrate optimization of bus routes and bus numbers in
consider various critical performance evaluation metrics. From the the route in order to streamline the working process.
observed data and functionality, preliminary analysis of indepen- UTK operates bus routes in and around the campus for the
dent and dependent variables is performed. In the transportation convenience of students, visitors, and physically challenged people
model, historical data, onsite observation of events happening and and faculty. The arrangement of buses is done under a contract with
customer feedback furnish a comprehensive idea of how things Knoxville Area Transit (KAT), Knoxville's city-owned public trans-
work in the real system. The most challenging task of the simula- portation provider. On an average, around 1.2 million passengers
tion process is transforming real life scenarios into system logic, are recorded as using the KAT-UTK service per year. The campus at
which is adaptable by the computer software. ARENA 10.0 simu- UTK itself is undergoing various infrastructure changes and
lation software was used for the simulation. ARENA is discrete development. These changes affect the transportation routes and
event simulation and automation software developed by Rockwell schedules, imposing the constant task of rerouting vehicles for
Automation. The processor and simulation language used by commuter convenience. The transportation department at UT
ARENA is SIMAN. operates six bus routes within the parameters of the campus to
The first step in achieving the dynamic simulation is short- transport students. The routes include the following:
listing the critical events that might happen over a long run with
a view to improving the performance metrics (the number of buses  The T EasteWest
in the route and average waiting time). A list of possible critical Weekdays only.
events is compiled, and proper actions are also prioritized and 7 a.m.e4 p.m. every 5e7 min
documented. The simulation model thus developed has the capa- 4 p.m.e6 p.m. every 10 min
bility to work through any change that might occur in the simula-  The T NortheSouth
tion model. The input to the system can now have a wide range of Weekdays only.
values to it, as the simulation model is now capable of recognizing 7 a.m.e3 p.m. every 5 min
the scenario and routing the entities through the best path for the 3 p.m.e6 p.m. every 10 min
most favorable solution.  T: Ag Campus Express

Fig. 1. Methodology to implement dynamic simulation.

436 G. Upreti et al. / Journal of Cleaner Production 85 (2014) 433e441

Table 1
Average number of passengers per day.

Ag express (Fall/Spring)

Avg. rider Avg. working days Avg. no. of riders

each month each month each day

Jan 3146 15 210

Feb 3988 20 199
Mar 3398 20 170
Apr 3632 20 182
May 213 4 53
Jun 0 0 0
Jul 0 0 0
Aug 2626 10 263
Sep 6080 20 304
Oct 4435 20 222
Nov 3673 19 193
Dec 722 7 103
Average 190

Fig. 2. Ag Express route map.

Weekdays only. night and also might cause inconvenience to physically challenged
8:45 a.m.e3:45 p.m. every 5 min students.
 T: Late Night
Sunday e Thursday, 6 p.m.e2 a.m. every 10 min 4.1. Base simulation model
Friday and Saturday, 6 p.m.e3:30 a.m. every 10 min
 T: Access The system that was used to evaluate the intelligent simulation
Weekdays only. model was the Ag Express route, which connects two ends of the
The T: Access is for persons with disabilities. The T: Access campus, transporting students attending classes and also serving
operates a point-to-point service on the UTK main campus, Ag the people who park cars in parking lots around the route. The
campus, or UTK facilities in the Ft. Sanders area, 7 a.m.e6 p.m. route was using four buses during any regular working day of Fall/
on demand. Spring semester, serving at an interval of 5 min. The software used
 T: Link to simulate this route was ARENA 10, which uses the SIMAN
The T: Link operates nightly between 6 p.m. and 7 a.m. The T: discrete event modeling technique.
Link transports students traveling alone at night to the T: Late The rationale behind the set parameters was that they would
Night or to their destination. The T: Link service area includes increase students' satisfaction in terms of transferring them be-
UT's main and Ag campuses, and Ft. Sanders to Grand Avenue. tween classes from one end of the campus to the other. The other
criterion that the express route is expected to perform is to trans-
Out of the six operating routes that were submitted to the team port students, faculty, and visitors who park in the main parking
for process improvement, only three routes were targeted for complex to their respective building areas. This function of the
improvement. No changes were suggested to T: Late Night, T: Ac- express route plays a major role in eliminating parking lot con-
cess or T: Link routes for process improvement, as the moral reason gestions. The express route is supported by a few other routes in
for the functioning of the route was solely the safety of the students and around the campus; as a system, the various parts of the T help
and process-efficiency measures that involved reducing the students travel faster and more safely around the campus to their
numbers of buses on special routes might risk student safety late at destinations. Fig. 2 illustrates the Ag Express route, displaying in

Fig. 3. Ag Express ridership history.

 G. Upreti et al. / Journal of Cleaner Production 85 (2014) 433e441 437

Fig. 4. Passengers entering the system.

detail the route, bus stops, the localities/buildings, and the parking the schedule prepared for the express route was more theoretical
lots that it covers. than real; i.e. the real system performance varied as follows:

4.2. Data collection  The time interval between each bus was not maintained at
5 min.
Historical and current data were collected and analyzed for the  Sometimes buses “stacked” or queued near a bus stop.
bus schedule, ridership data, driver schedule, bus timing, and fre-  Stops were made whenever any person gestured to the driver of
quency. The ridership of students varies significantly throughout the bus to stop.
the year, as shown in Fig. 3.  Unscheduled breaks were taken by the drivers.
A number of onsite surveys were conducted to collect data  The buses were rarely filled, and a few instances of buses
relevant for developing the simulation models. Upon conclusion of running empty were also documented.
the survey process, the team concluded that there are many more
passengers en route between class hours, and during morning and Using the provided, collected, and surveyed data, the simulation
evening hours. The number of occupants in the buses is minimal model was created in ARENA platform. The base model consisted of
the rest of the time. It is noteworthy that, even during peak hours, four buses, each starting one after the other at regular intervals
the buses seldom carry enough passengers to reach the buses' from the same main station.
seating limit. Table 1 lists the annual average number of riders,
average working days in a month, and average riders each month, 4.2.2. Base model development
with average riders per day derived from those data. Simulation is the process of replicating the real working
condition of a system using computer logic. Every simulation
4.2.1. Initial data analysis model is unique, as the model logic depends upon the thinking
Ag Express is supposed to operate every 5 min from 8:45 a.m. process of a simulating team or of individuals. This simulation
until 3:45 p.m. On-site investigation of the bus route reveals that model was created by considering both passengers and buses as

Fig. 5. Passengers exiting system.

438 G. Upreti et al. / Journal of Cleaner Production 85 (2014) 433e441

Fig. 6. Base simulation model.

Table 2
Result from the bases simulation model. also shows that the process involves a series of divide and separate
modules.
Avg. passenger waiting 4 resources 3 resources 2 resources
After the destined entities exit the system, the remaining
time (mins)
passenger entities along with the new passenger entities are
Bus stop 1 2.594 3.821 6.281
clubbed again into the bus and the process is repeated. The ca-
Bus stop 2 2.601 3.829 6.163
Bus stop 3 2.715 3.896 6.132 pacity of each bus is 30 and the system logic does not allow more
Bus stop 4 4.094 6.564 6.400 than 30 passenger entities to be batched with a single bus entity.
This is achieved using a series of IF statements in the Hold
modules. Fig. 6 shows the actual simulation model with a few
entities. A series of combined and separate blocks was used to blocks being hidden in sub modules. The actual performance of
replicate the logic of passengers entering and exiting buses. Each any simulation model depends on its proximity to the real sys-
passenger and bus entity was given individual attributes to tem performance. Therefore, verifying the simulation model is as
identify them throughout the system. Variables were used important as developing the simulation model.
to control metrics such as the bus's seating capacity, the number Fig. 6 shows the collective bus traveling route, the passenger
of people entering the system, the number of people exiting the boarding area, the passenger exiting area, and the performance
system, and the number of passengers currently on the bus. metrics, which are highlighted using graphs. The base simulation
Fig. 4 shows the arrangement of passengers entering the system. model is developed to simulate a day's work in one run. Unlike
All the passengers were identified according to their respective manufacturing operations where unfinished parts are left in the
destination, and a batch module in ARENA was used to merge each system to be picked up when the next shift starts, the trans-
bus with the arriving buses. Each bus, after getting batched, trav- portation system re-organizes itself to a fresh start every shift.
eled to other bus stops. At the bus stops, each bus was separated There are no passengers left in the system at the end of the day. The
and the passengers whose destination matched to the bus stop transportation model is run in three sets with 100 replications for
exited the bus. Fig. 5 illustrates the exiting process of the bus stop. It each set. For each set, the number of resources is reduced one at a

Table 3
Savings from the base simulation model.

Row Route Current Hours/day Current cost Price($) Suggested Hours/day Savings($) Avg. Regular Total savings
resources used ($)/hour /day resources used /day days/year ($)

1 EasteWest 7 68.5 42 2877.0 6 61.2 315 158 49,770

2 Late Night 2 21.2 42 890.4 2 15.83 199.5 134 11,256
3 Ag Express 4 30.7 42 1289.4 2 15.33 644 155 99,820
4 NortheSouth 2 20.5 42 861.0 1 11.67 364 155 56,420
Total 217,266
 G. Upreti et al. / Journal of Cleaner Production 85 (2014) 433e441 439

Table 4 The scenarios which have been kept in this model actually
Critical performance metrics in transportation model. replicate executive decisions in case of crisis. Assumptions include:
Critical metrics Critical events Deliverables

No. of people waiting Demand of students Resource

 The number of bus stops and buses do not exceed more than six
Time of day allocation at any given time.
Avg. waiting time of people Day of week Routing of  Depending upon the average waiting time of passengers at each
Special events buses bus top, the buses are provided specifically at places where they
Parking facilities
are needed the most.
 When additional buses are needed, the task of the bus is to
deliver onboard passengers to their destinations and then to exit
time to estimate the average waiting time of passengers at each bus the system to the bus depot.
stop (see Table 2).  The model is run for four varying schedule types termed in the
The results of the simulation model are the following. model as high frequency schedule, low frequency schedule,
medium frequency schedule and schedule with gaps in
 The average waiting time of the passengers for the bus increases between.
from around 2.59 min when four buses are used to 3.8 min with  The capacity of the bus is limited to 30 passengers.
three buses, to 6.4 min for two buses.  The boarding time of passengers is assumed to be negligible.
 Even with just two buses serving the route, the number of  There are two different routes for the Ag Express. The shortest
people in a bus is never more than 12 on an average working one has four bus stops and longest one has six bus stops.
day.
The critical events are set in the validated base simulation model
Table 3 shows the calculated cost savings projected by the using the following logic cases.
transportation project. Considering only the Ag Express route, per
annum cost savings of $99,820 are possible to achieve simply by  The Clock Model
reducing the number of buses from four to two.
A clock model employs loop simulation and is a part of the base
simulation model. It has its own seconds, minutes, and hour vari-
4.3. Dynamic simulation modeling ables, which are used to control the main simulation model. Fig. 7
details the clock simulation model and its ARENA modules.
The purpose of the dynamic simulation model is to build a
simulation model which will adapt itself to the changing conditions  Dynamic Assignment of Buses
and which is more accurate in its construction. In a traditional
simulation model, changes have to be made when critical events One of the features of the dynamic simulation model is the
happen in the real system. Dynamic simulation modeling mini- addition of buses to the specific bus stop when the average waiting
mizes the effort needed for timely updating the simulation model time of passengers is higher than normal. This addition is achieved by
as all the critical events are added at the development stage calculating the average number of passengers in the hold module of
increasing the reliability of the simulation model. The dynamic the bus stop and using this average number as a condition to release
simulation model construction begins with documenting the list of buses into the system.
critical events. Table 4 below lists the critical performance metrics Another loop is created in order to remove buses from the system
in the case study model. if there are fewer passengers to satisfy. This feature is also coded

Fig. 7. Clock model.

440 G. Upreti et al. / Journal of Cleaner Production 85 (2014) 433e441

Table 5
Results of dynamic simulation model.

Replication Expected no. of Actual no. of No. of Entities Event No. of No. of buses
no. people entering people entering people per stops
Incre- Decre- Avg. waiting time Avg. no. of
exiting arrival
menting menting of people (mins) people waiting

1 35 30 30 1 Summer 4 1 0 13.492 0.105

2 52.5 33 33 1.5 Summer 4 1 0 15.522 0.131
3 70 84 84 2 Summer 4 5 0 17.901 0.366
4 101.5 80 80 2.9 Fall/Spring 4 1 0 15.356 0.309
5 129.5 123 123 3.7 Fall/Spring 4 2 1 14.784 0.453
6 161 172 172 4.6 Fall/Spring 4 5 3 12.824 0.543
7 182 185 185 5.2 Fall/Spring 4 8 8 12.268 0.565
8 210 228 228 6 Fall/Spring 4 6 5 12.552 0.716
9 315 270 270 9 Game day 6 7 1 12.588 0.866
10 385 385 385 11 Game day 6 11 5 13.494 1.286

Table 6
Time of entry and exit of buses.

Run Time Bus no.

number (mins)

Bus entering the system 5 0 1

5 100.550 2
Bus exiting the system 5 150.395 2

using the same average number of passengers waiting in the bus stop.
A “decide” block looks for the average number of passengers waiting
at the bus stop and channels it into a parallel loop, where the buses
are taken around other bus stops to deliver passengers to their des-
tinations. Once the number of passengers in the bus is zero, the bus
exits the system. The buses can be pulled into the system and out of
the system any number of times based on this requirement.

 Dynamic Travel Times

The ideal travel time between any two consecutive bus stops is
5 min, as per UTK's Parking and Transit Services Department. But
traffic density changes in and around university roads according to Fig. 8. Results from ANOVA.
the time of day. In the dynamic simulation model, the time vari-
ables in the clock model are used to generate three different travel
In the dynamic simulation model, the incoming number of
times for buses depending on the time of day.
passengers is read by the dynamic simulation model using a
series of “Read and Write” modules. For each run, the incoming
 Event and Route Decisions
number of passengers changes dynamically, triggering a new
event and route for each run.
Traditionally, the purpose of the express route is to connect
the two ends of the UTK campus and to get people from parking
lots to their destinations. Events like football games, basketball 5. Results and discussion
matches and university festivals draw huge numbers of people to
the campus. On occasions like these, two extra university-owned This simulation case study illustrates the possibility of using
parking lots on the far edges of campus are used in order to park dynamic simulation methodology to develop a self-adapting
cars, and the buses are used to transport people into the main simulation model that changes itself over various critical
campus.
Table 8
Comparison between real, traditional and dynamic model.

Table 7 Current Suggested Current Price ($) Annual

Cost of operation calculation for dynamic model. resources resources cost /day operating
($)/hr cost ($)
Bus Start time End time Hours of operation Cost of
Present 4 0 42 1289.40 199,857.00
(mins) (mins) (mins) operation ($)
model
1 0 500.00 500.00 350.00 Traditional 0 2 644 155.00 99,820.00
2 93.560 154.071 60.511 42.36 simulation
3 124.066 168.796 44.730 31.31 model
4 207.081 305.057 97.976 68.58 Dynamic 0 5 731 155.00 113,235.12
5 404.860 500 95.140 66.60 simulation
Total 798.358 558.85 model
 G. Upreti et al. / Journal of Cleaner Production 85 (2014) 433e441 441

conditions. In a traditional simulation model, for each repetition, the base simulation model, and the result is obtained, with the
random values are taken within the specified values, and the exact number of resources needed, the hour of need, and the time
average of all the runs are used to generate results. In a dynamic frame of need all identified. In the traditional simulation model, the
simulation model, for each run a controlled input data is assigned number of buses is allocated manually and buses are operating in
that will trigger a series of events specific to that one particular the route even when not required.
condition. The path which the model assigns itself is the result of Table 8 is a visual comparison between the operating cost of the
the simulation model. Thus, in a dynamic simulation model, all real model, the traditional simulation model, and the dynamic
possible conditions that occur and events that might occur together simulation model proposed by this research. The result of the dy-
can be replicated. The case study provided is a brief illustration to namic simulation model suggests using four buses for the shift with
demonstrate the feasibility of the dynamic simulation three buses returning to the depot when not needed. The dynamic
methodology. simulation model sends and removes buses from the system ac-
The actual number of passengers entering the system will be an cording to demand, bringing down the buses' usage from 30.7 to
approximation of the expected number. The number of bus stops 13.31 h per day. The reduction in number of buses also reduces the
and the route is decided by the simulation model and depends pollution caused by emissions. The proposed model can be applied
upon the incoming flow of passengers. Depending upon waiting to transportation systems for any organization. The projected
times of passengers at the bus stops, buses are released one at a annual cost savings using the dynamic simulation model is
time to the exact bus stop of need. The bus number, time of release $113,235.12. This is equivalent to an annual saving of 28,309 gallons
and the simulation run number are displayed in the simulation of diesel, avoiding the emission of 628,460 lbs of carbon dioxide
result. In a similar manner, the buses are removed from the route into the environment.
when the average waiting time of passengers goes down drastically.
The simulation's logic code ensures that when a bus is selected to
References
depart from the system, the passengers on the bus reach their
destinations. Banks, J., 1999. Introduction to simulation. In: Farrington, P.A., Nembhard, H.B.,
Table 5 shows the series of decisions the simulation model had Sturrock, D.T., Evans, G.W. (Eds.), Proceedings of the 31st Conference on Winter
to make for different passenger input rates, different runs and Simulation: Simulation—a Bridge to the Future, vol. 1, pp. 7e13.
Banks, J., Nelson, B., Nicol, D., 2009. Discrete-event System Simulation. Prentice Hall.
number of passengers entering. The template can be expanded to Barnes, P., Jerman, P., 2002. Developing an environmental management system for a
show details regarding individual runs, such as the average waiting multiple- university consortium. J. Clean. Prod. 10, 33e39.
time of passengers at each bus stop, the release time of an indi- Chick, S.E., 2001. Input distribution selection for simulation experiments: ac-
counting for input uncertainty. Oper. Res. 49 (5), 744e758.
vidual bus into the system and the exit of the individual bus from El-Haik, B., Al-Aomar, R., 2006. Simulation-based Lean Six-sigma and Design for Six-
the system. sigma. http://encompass.library.cornell.edu/cgibin/checkIP.cgi?access¼gateway_
An example of the dynamic simulation model is illustrated in standard%26url. http://www.books24x7.com/marc.asp?bookid¼16837.
Fishman, G.S., Kiviat, P.J., 1967. The analysis of simulation-generated time series.
Table 6, where for “Replication 5” it is shown that two buses Manag. Sci. 13 (7), 525e557.
entered the route to meet the demand and one bus exited the Greene, D.L., Duleep, K.G., Upreti, G., 2011. Status and Outlook for the U.S. Non-
system in between the simulation run. Detailed metrics illustrating automotive Fuel Cell Industry: Impacts of Government Policies and Assess-
ment of Future Opportunities. U.S. Department of Energy.
the time of bus release and time of recall can also be traced using Kelton, W.D., 1997. Statistical analysis of simulation output. In: Andrado ttir, S.,
the set methodology. Healy, K.J., Withers, D.H., Nelson, B.L. (Eds.), Proceedings of the 29th Conference
The base simulation model is compared to the dynamic simu- on Winter Simulation, pp. 23e30.
Kelton, W., Sadowski, R., Swets, N., 2009. Simulation with Arena. McGraw-Hill
lation model for its effectiveness in obtaining and understanding
Higher Education.
results. To compare the dynamic simulation model with the Lozano, R., 2004. A Tool for Easy Benchmarking Sustainability Reports in Univer-
traditional simulation model, a comparable situation is applied, sities. Environmental Management Sustainable Universities Monterrey, Mexico.
namely the number of passengers entering the system and the Manuj, I., Mentzer, J.T., Bowers, M.R., 2009. Improving the rigor of discrete event
simulation in logistics and supply chain research. Int. J. Phys. Distrib. Logist.
number of bus stops. The result of this simulation is compared to Manag. 39 (3), 172e201.
the base simulation model to calculate the effective savings. Table 7 Ming, R., Jiang, T.S., Tsai, J.J.P., 1990. Integrated intelligent simulation environment.
and Fig. 8 show the analysis of dynamic allocation of buses over Simulation 54 (6), 291e295.
Robinson, S., 2005. Discrete-event simulation: from the pioneers to the present,
time. The primary advantage of having a dynamic simulation model what next? J. Oper. Res. Soc. 56 (6), 619e629.
capable of taking decisions based on set constraints is that it per- Robinson, S., Edwards, J.S., Yongfa, W., 1998. An expert systems approach to
mits a dynamic allocation of buses during hours of need. simulating the human decision maker. In: Medeiros, D.J., Watson, E.F.,
Carson, J.S., Manivannan, M.S. (Eds.), Proceedings of the 30th Conference on
Results from ANOVA test shows that the waiting time for the Winter Simulation, vol. 2, pp. 1541e1545 (Washington DC).
dynamic model is higher than the base model. Even though the Robinson, S., Alifantis, T., Hurrion, R., Ladbrook, J., Edwards, J., Waller, T., 2001.
waiting time is high, cost saving is significant as shown in Table 8. Modelling and improving human decision making with simulation. In:
Peters, B.A., Smith, J.S., Medeiros, D.J., Rohrer, M.W. (Eds.), Proceedings of the
Also, small p-value suggests that the model is statistically
33rd Conference on Winter Simulation, vol. 2, pp. 913e920 (Arlington, VA).
significant. Savely, S.M., Carson, A.I., Delclos, G.L., 2007. An environmental management system
Table 8 shows the effectiveness of the base simulation model in implementation model for U.S. colleges and universities. J. Clean. Prod. 15,
660e670.
terms of suggestions proposed. The traditional simulation model
Upreti, G., Greene, D.L., Duleep, K.G., Sawhney, R., 2012. Fuel cells for non-
employs reducing the number of buses, manually, one at a time, in automotive uses: status and prospects. Int. J. Hydrog. Energy 37 (8), 6339e6348.
order to generate results. Each result with a different number Velazquez, L., Munguia, N., Platt, A., Taddei, J., 2006. Sustainable university: what
of resources is compared to decide on the optimal number of re- can be the matter? J. Clean. Prod. 14 (9), 810e819.
Wagner, M.A.F., Wilson, J.R., 1995. Graphical interactive simulation input modeling
sources needed. In the dynamic simulation model, more environ- with bivariate Bezier distributions. ACM Trans. Model. Comput. Simul. 5 (3),
mental conditions and constraints are considered than they are in 163e189.