Reducing Truck Emissions in Import Operations at Container Terminal—A Case Study in a Singaporean Port - Technology Management for Sustainable Production and Logistics - page 133

Truck turn time is defined as the duration from the arrival of a truck at the
terminal gate to the moment of exit. Existing studies identify three modeling
approaches to estimate truck turn time and its components: (a) simulation mod-
els (Huynh et al. 2004 ; Huynh 2009 ), (b) regression models (Huynh et al. 2004 ;
Goodchild and Mohan 2008 ), and (c) queuing models (Guan and Liu 2009 ; Chen
et al. 2011a , b ).
Huynh et al. ( 2004 ) develop a discrete event simulation model of a container
terminal, representing the precise movements of trucks and yard cranes. The simu-
lation model is used to find the number of yard cranes needed to achieve a desired
truck turn time. Huynh ( 2009 ) developed a simulation model to evaluate perfor-
mance of various rules for truck arrival management. Both simulation models are
powerful tools to present detailed operations of trucks and terminal equipment and
can be used to test different operation scenarios. On the other hand, simulation
models require long computational times, as a large number of replications are
needed to reduce sampling variance and provide reliable results. This is a major
barrier for integrating a simulation model into an optimization process. In this
study, one of the contributions is to improve in this aspect by programming both
the simulation model and the optimization model on the same platform, so at to
improve the algorithm computational efficiency.
Huynh et al. ( 2004 ) applied regression analysis on the output of their simula-
tion model for different scenarios of container terminal operations. A second order
polynomial function was developed to predict truck turn times, with an adjusted
R 2 of 0.7381, using the average truck number served by a crane as the predictor
variable. Goodchild and Mohan ( 2008 ) discussed the prediction accuracy of the
model by Huynh et al. ( 2004 ) and claimed that using averages rather than a single
simulation replication reduces the variability and improves the fit of the regression
model (i.e. high R 2 value). To support their claim, they developed a linear function
of truck turn times using a single simulation replication data of truck arrivals. The
R 2 of their model was only 0.1709. Therefore, regression models are not accurate
enough to estimate truck turn times at terminals.
There are two types of queueing models used to estimate truck turn times: (a)
conventional stationary queueing models, and (b) non-stationary queueing mod-
els. Guan and Liu ( 2009 ) analyzed truck queues at marine container terminal gates
with a stationary M / E k / c queueing model. A limitation of stationary queueing
models is they neglect the transient behavior and only analyze the steady state of
a queue. This raises concerns about the applicability of simple stationary models
(Green and Kolesar 1991 ). Typically, truck queues at marine container terminal
gates are not in a steady state, as truck arrival and gate service rates vary over
time. This indicates the need to use state-dependent queueing models, as they are
more effective and robust tools to capture time-varying behavior of such queue-
ing processes (Smith 2010 ). In a pioneering study by Green and Kolesar ( 1991 ), a
point-wise stationary approximation (PSA) was proposed to model non-stationary
queueing systems. Whitt ( 1991 ) further verified that the PSA model is asymp-
totically correct as the service and arrival rates increase with fixed instantaneous
traffic intensity. Wang et al. ( 1996 ) proposed a point-wise stationary fluid flow