Jurnal
Teknologi
Full paper
Univariate Throughput Forecasting Models on Container Terminal
Equipment Planning
Jonathan Yong Chung Ee*, Abd Saman Abd Kader, Zamani Ahmad, Loke Keng Beng
Department of Mechanical Engineering, Universiti Teknologi Malaysia, 81300 UTM Johor Bahru, Johor, Malaysia
*Corresponding author: jonathantbyce@yahoo.com
Article history
Abstract
Received :27 March 2014
Received in revised form :
10 April 2014
Accepted :5 June 2014
Planning of Container Terminal equipment has always been uncertain due to seasonal and fluctuating
throughput demand, along with factors of delay in operation, breakdown and maintenance. Many timeseries models have been developed to forecast the unforeseen future of container throughput to project the
needed amount of port equipments for optimum operation. Conventionally, a "ratio" method developed
by port consultants at early port design stage is adopted for equipment planning, giving no consideration
to the dynamic growth of the port in terms of improved layout and technological advancement in
equipments. This study seeks first to enhance the empirical approach of the equipment planning at the end
of planning time horizon by including assumed coefficient of port capacity parameters. The second is to
compare the size of equipment purchase by receiving different terminal's future throughput demand from
two univariate forecasting models at planning time horizon. The empirical method of equipment planning
will be tested against the conventional yard equipment per quay crane ratio after deriving the throughput
demand from forecasting models of Holt-Winter's exponential smoothing and seasonal ARIMA
(autoregression integrated moving average) model. Results in the form of graphs and tables indicate
similar forecasting pattern by two models and equipment estimation proofs to avail more redundancy for
optimum operation. Suggestions for better estimation of equipments are also made for future models.
Graphical abstract
TEU Throughput Forecast
9E5
8E5
7E5
6E5
5E5
4E5
3E5
2E5
1E5
1995
1997
1999
2001
2003
2005
2007
2009
2011
2013
2015
2017
TEU Throughput Forecast Projection
Keywords: Port planning; equipment forecasting; univariate; ARIMA; exponential smoothing
© 2014 Penerbit UTM Press. All rights reserved.
1.0 INTRODUCTION
Containerization has been an important key in the rapid growth of
international trade, particularly for Malaysia as a strategic midway-point of the east and west. So, demand has been high on
terminals logistics, management and technological breakthrough
to be able to service the immense growth of container shipment
[1], transshipment or inbound container service. Malaysian ports
capacity is one of the most important determinants of meeting the
increasing trade.
Port capacity refers to the ability of the port to provide a
minimum queuing time, berthing and efficient operation of
handling container transfer. An optimum capacity can be defined
by various approaches such as economical, theoretical, empirical,
operational, engineering or integrated [2].
Container terminals planning is a dynamic system with
multiple-aspect consideration to fit for an optimum operation the
staggering array of container handling equipment (CHE) that is
available. In general, CHE can be categorized into fixed
equipments (container conveyer, automated stackers and container
lifts), rail-mounted equipments (ship-to-shore cranes, gantries,
transtainers and trains) and free-running equipments (rubber-tiregantry, forklifts, reachstakers, straddle carriers and primemovers). Inability to match equipment capacity to the handling of
throughput demand may result in long queuing time long after the
estimate time of arrival (ETA) and possible hiccups in the loading
and unloading process of container from vessel to the respective
storage area [3]. Failure to provide service at ETA will impose
additional cost and time that may devastate the port's reputation
and drive ship-liners away to the nearest adjacent port for a more
swift and reliable service. Not only does this cause irreversible
investment loss but also the withdrawal of many stakeholders for
future investment. [4].
Therefore, substantial planning for additional port
equipments, whilst not giving high redundancy that incur high
investment, is of crucial importance for optimum operation of
container movement [5]. Traditionally, the number of freerunning equipments required is selected based on a crane/berth
ratio. This ratio is predetermined by port consultants [6] at early
port establishment as a function of the specific layout of the port.
This ratio is non-sustainable as the port expands and cannot stand
as a optimum equipment ratio at all time. There is an obvious lack
of general empirical estimation to account for the current
crane/berth ratio to execute optimum operation in the container
transfer process.
In order to plan for equipments as well as port expansion,
prior long term forecast of container throughput is an essential
reference to gauge the throughput growth size .Monitoring
69:7 (2014) 163–171 | www.jurnalteknologi.utm.my | eISSN 2180–3722 |
164
Jonathan Yong Chung Ee et al. / Jurnal Teknologi (Sciences & Engineering) 69:7 (2014), 163–171
throughput changes of seasonal growth pattern aides planning
activity involving the acquisition of additional equipment and
even arrangement of the port's transportation system. However,
permutation of transport system is not the focus point of this
paper. This paper highlights the activity of equipment acquisition
based on the forecasted throughput demand at the end of the longterm planning, also known as planning time horizon (PTH). There
are a vast discipline of forecasting techniques that include timeseries projection, simulation of one or more variables, inputoutput analysis and qualitative forecasting by expert opinion [7].
As of this paper, time-series projection is chosen as a forecasting
tool for equipment planning – a total empirical approach.
In short, this paper has a twofold objective, first, to enhance
the existing empirical formula for not only the number of quay
crane needed but also to for other major equipments namely RTG
and prime-movers. This follows the previous work on equipment
estimation [2], [8], [9] with differing parameters and
consideration to amend the existing algorithms. Second, to
investigate the influence of different forecasting models [10],
[11], [12] projecting the required number of equipments for
optimum operation. The methodology adopted is to compare the
significant difference of the enhanced empirical method to
conventional equipment estimation methods from a set of sample
data (undisclosed port name) by forecasting the future throughput
from Holt-Winters method and seasonal ARIMA (0,1,1)(0,1,1)12 –
Box-Jenkins method. The results will be analyzed for accuracy
and significant increase in numbers of equipment required at the
end of the planning time horizon, set to be 5 years.
2.0
LITERATURE REVIEW ON EQUIPMENT
ESTIMATION
APPROACH
AND
FORECASTING
METHODS
2.1 UNCTAD and Other Empirical Models
One of the first reference for port expansion points to the
UNCTAD model [13]. Equipment expansion is one of the four
elements in UNCTAD development and expansion model, others
being the container park area, container freight station and berthday requirement. The determined number of container handled per
crane, number of cranes per ship and moves are in function of perberth requirement. UNCTAD model provides no empirical
formula except but a simplistic graphical chart to estimate future
demand support based on the four variables mentioned. However,
it lacks control in terms of uncertainty and has a short range of
parameter which cannot accommodate designs of larger terminal
size. While others who followed after UNTCAD came up with
various numerical models to analyze the handling capacity
demands [7], [14], [15], [16], [17], [18], [47]. Loke (2012)
equipment estimation formula is as below:
∆
,
2.2 Costing Models
ℎ
= (
� �.
∆
.
ℎ
.
)
From the economical-feasibility stand-point, costing is also a
basis of determination of equipment procurement size that became
the focal point of researchers [19], [20]. Dekker's model for
equipment expansion is based on a marginal approach that
analyzes the need of expansion at intervals according to the
current capacity of the terminal [8]. The basic approach is to
calculate the optimal expansion by economical-order-purchase at
steady-state-demand growth of each equipment [21]. Cost model
provides specific terminal equipment expansion with relative
control in the financial costing of equipment procurement.
2.3 Queuing Theory
Queuing theory is a time definition of an entrance units queue in
an immediate service of unloading or loading containers at the
container berths and leave the system when the service has been
performed [22]. Queuing theory holds on to the criteria that it is
possible to accurately predict servicing time of ships rather than
the estimated time of arrival of ships at terminal. Therefore, the
objective is to ensure the handling capacity is equal or greater
than the number of arriving ships, and so estimate the required
number of port equipment. Consequently, queuing theory
expressed itself in the form of berth occupancy rate .
1 tarr , ( t arr being the consecutive ship arrivals) and
1 tserv ( t serv being the reciprocal value of the service rate)
Where,
Based on parameters of consideration, the queuing problems
are solved by iterating parameters of users preference such as
probability of occupied berth, at service or unoccupied; average
time of queue, time of service, number of ship queue, etc.
Changes in values of terminal indexes and its impact on other
parameters are largely computed by using Poisson's distribution.
Advance simulation language based on queuing theories are such
as PORTSIM [23], Modsim III [24],, SIMPLE++ [25], ARENA
and SLX [26], Visual SLAM [27], AweSim [28], etc.
Queuing theory not only models the required port parameters
for sufficient support for optimum operation, it also provides a
basic model for queuing cost which is of great interest optimize
the service demand whilst not oversupplying equipment which
leads to an uneconomical operation.
2.4 Conventional Way & Practices By Industry
By common practice, port authority's planning for equipment size
is in accordance to a predetermined yard equipment to ship-toshore crane designed by port consultant firms [6] in the early
stage of the port development. This ratio is in function of the
yard's physical layout, therefore, the ratio value differ for each
distinct port due to their varying design of layout–parallel by RTG
terminal or perpendicular by RMG terminal [29], and the
demanded service rate. The assumption of this ratio estimation is
that the horizontal transport capacity must be at least equal to the
maximum quay handling capacity.
2.5 Forecasting Methods
Frankel [7] elaborates a series of forecasting methods in
predicting container throughput demand that includes prospective
economic over a time period of interest and other development
such as the economical and social effect, specified port,
modernization of existing port facilities, maintenance and
investment implications.
In response to varying aspects of forecasting influence,
forecasting projections are often done with the input of at least
one input of data to multiple sources of input. Even so, forecasting
can be classified into several approaches for example, model
building and simulation, qualitative forecast and time series
projection [7].
165
Jonathan Yong Chung Ee et al. / Jurnal Teknologi (Sciences & Engineering) 69:7 (2014), 163–171
Historical data is the most reliable source of data to be interpreted
for forecasting purposes. Models and simulation processes the
data by techniques of trend extrapolation, pattern and probabilistic
forecasting producing arrays of forecasts and decision tree
matrices. Numerous algorithms and formulas developed are
combined and adopt multivariate regression model [30]
assumption to iterate the interrelationship between sets of
variables. The relationship is then used to forecast the future
throughput by fitting the characteristic observed by the model. An
attribute theory, a new artificial intelligence [31], developed a
fourfold attribute for throughput prediction. The influence of
GDP, cargo throughput, foreign trade and total import and export
volume are amongst the relevant analysis with distinct pattern that
when merged give a multi-spectrum of forecast of throughput
which can be more reliable [32].
For qualitative forecasting, it deals with trades which are not
susceptible to extrapolation and analysis or any methods that
heavily builds on existing data. Expert surveys are main source of
information to predicting future throughput with their rich
experience as technologist, operators, planners and others.
Table 1 Previous researches scope on unvariate forecasting for container throughput
Authors
Subject of Forecast
Model of Univariate Forecasting
Li, X. and Xu,
S. (2011) [32]
Container Throughput
- Shanghai Port
-Cubic exponential smoothing,
-Grey Model,
-Multiple regression analysis
Xie, G., et al.
(2013) [34]
Container Throughput
– Hybrid Approach
- SARIMA model
- X-12 ARIMA model
- Classical Decomposition Model
Chen, 2010
[35]
Container Throughput
– Comparative Study
Chou, C.-C.,
et al. (2008)
[36]
Seabrooke,
W., et al.
(2003)
[37]
Peng, W.-Y.
and C.-W.
Chu (2009)
[38]
Gosasang, V.,
et al. (2011)
[39]
DuanXueyan,
XuGuanglin,
Yu Siqin
(2012)
[31]
Technique
Optimization by dual
combined technique
Findings
Hybrid combination with
"least quares support
vector regression" model
Highlighting the importance of
capturing seasonal and non-linearity
for better forecasting
- Genetic Programming
- ARIMA (X-11)
- SARIMA
Comparison of few
approaches
Suggesting Genetic Programming as
optimum forecasting method
Container Throughput
– Taiwan Port
- Regression model
Generate a new modified
regression model
Results proposed modified
regression model for higher
prediction accuracy
Container Throughput
– Hong Kong Port
- Conventional Regression model
Forecasting with other
affecting factors.
Results yield more reliable pattern
of throughput growth
Container Throughput
– Comparative Study
- Decomposition
- Trigonometric
regression
- Regression with seasonal dummy
- Hybrid grey model,
- Sarima model
By applying monthly data
input and evaluation of
error profile
classical decomposition
model appears to be the best model
for forecasting container throughput
with seasonal
variations.
Container Throughput
– Comparative Study
- Neural Network
- Linear Regression
By measurement of
RMSE, MAE
Neural Network as being the best
application for forecasting
Container Throughput
– Application Study
Attribute Theory (artificial
intelligence forecasting)
By applying mapping
theory and conversion
degree function
Attribute theory is effective and
feasible with comprehensive
consideration of influencing factors
Delphi method [33] is one of the popular approach by welldefining questionnaires to specific parties that yields a consensus
of factors and opinions on future container throughput.
In the spectrum of time-series forecast, also called the
univariate forecasting model, is particularly useful when little is
known about the underlying revolution of the history of container
throughput pattern. From a simple linear regression extrapolation
to neural network analysis [40], time-series forecasting models
has become complex and detail in finding the most fitting
function to the real variable in order to forecast according from
past trends and pattern. Researchers today developed better
forecasting models by comparing with other models through
varies application and techniques like ACF, PACF, MAPE, etc.
Table 1 shows the recent researches and its findings of superior
performing models for specific applications.
The basic forecasting approach are mostly regression-based
and highlights the importance of placing weightage on seasonal
pattern consideration and the analysis of error (RMSE, MAE,
MAPE).
Exponential smoothing and
regression model prove better
prediction
2.6 Holts-Winter's Exponential Smoothing
Some of the most successful forecasting methods are based on the
concept of exponential smoothing. Exponential smoothing
techniques are simple tools for smoothing and forecasting a time
series that is, a sequence of measurements of a variable observed
at equidistant points in time. Smoothing time series aims at
eliminating the irrelevant noise and extracting the general path
followed by the series. The first exponential smoothing originated
from the work in the US Navy in 1944, Robert G. Brown
developed the algorithm when tracking the velocity ad angle used
in firing at submarines as a research analyst. He further developed
it in 1950s to discrete time series handling trend and seasonality
[10]. Later Charles Holts in the US office of the Naval Research,
developed a different smoothing trend and seasonal component
with the novelty of incorporating the additive ad multiplicative
component [11]. Soon, his student Peter Winters [42] provided
empirical test on Holts method and, therefore, the seasonal
version of Holt's method is called Holt-Winters' method.
There are two variations to this method which differ is in the
seasonal component. The additive method is preferred when the
seasonal variations are roughly constant through the series, while
166
Jonathan Yong Chung Ee et al. / Jurnal Teknologi (Sciences & Engineering) 69:7 (2014), 163–171
the multiplicative method is preferred when the seasonal variations are changing proportional to the level of the series [43].
total number of years, that is set as planning time horizon as
below:
T = ( T ℎ)
T ℎ = years
2.7 Seasonal ARIMA (SARIMA)
The AutoRegressive Integrated Moving-Average (ARIMA)
procedure analyzes and forecasts equally spaced univariate time
series data, transfer function data, and intervention data. This
univariate time series model was first popularized by George Box
and GwilymMeirion Jerkins, frequently called “Box–Jenkins
models” because it [12] proposed a complete methodology for
modeling time series which consists of three phases:
identification, estimation and testing, and application. It is suitable
for use with a stationary time series. However, ARIMA and
SARIMA are built on linear assumptions and they cannot capture
the nonlinear patterns hidden in the original data, which leads to
poor forecasting performance [41].
3.0 METHODOLOGY
This section explains the framework of calculating and comparing
the outcome from forecast data derived from the Holts-Winter and
SARIMA method. The empirical method of calculation is tabled
against the conventional equipment ratio estimation, whilst the
actual planning figures from port data will be compared. The
empirical calculation is done by adopting the parameters like
equipment capacity, coefficient and handling rates [6], [7], [39]
and equipment profiles. Equipment ratio estimation is simply
taken from planning practice figures to extrapolate equipment size
by number of projected throughput. Figure 1 below is a general
flow-chart describing the framework.
To match the capacity of the various equipments in support
of the increased throughput, the following equation determines the
additional number of equipments need for the new expansion
phase.
∆
∆
SARIMA
ni,pth
ni,o
∆n i,pth
Conventional
Equipment Ratio
Estimate
Equipment Empirical
Calculation
Results of Estimated
Equipments
Comparative
Analysis
Figure 1 The overall process of the comparative forecasting model for
equipment estimation
3.1 Equipment Estimation
The equipment planning time frame selected here is 5 years,
which is a common practice in port expansion. Tt represents the
–
---(1)
, −
,
ℎ
=
,
ℎ–
. .
.
---(2)
,
Point of interest after forecasting the throughput is the
number of equipment, which is calculated empirically by the
following equation (UNCTAD, 1976; Loke et al., 2004). The
basic algorithm by Loke is enhanced by including additional
factors from earlier reference, TEU factor, handling ratio,
maintenance period and the breakdown of equipments move per
hour (Equation 3) is added.
,
Comparative
Analysis
,
∆ ni , n i,j , n i,j-1 represents the additional equipment needed,
number of equipment needed at expansion, number of equipment
available since previous expansion. Function i is numbered by
1,2,3 as 1= qc (quay crane); 2 = rtg (rubber-tire gantry); 3 = pm (
prime-mover); j represents the expansion year phase (j = 1, 2,
3,…, pth). To derive the number of equipments need for the entire
planning time horizon (pth), the equation (1) is adjusted by setting
in the planning time phase and the initial available number of
equipments.
Time Series Data
Port TEU History
Holts-Winter
Exponential
Smoothing
=
,
Qpth
f
MPH i
Ts
nb
ri
uti
mdt
ℎ
∆
= (
� .� � .
,
ℎ
=
�.
�� .
ℎ
. .
��ℎ
�
.
) (
. −
�
−
–
,
)
---(3)
---(4)
=total number of equipment i, at the end of planning
time horizon
= number for addition equipment i at initial planning
phase t=0
= number for addition equipment i, at planning time
horizon from planning time, t=5
= throughput amount at the end of planning time
horizon (TEU)
=TEU factor
= Moves per hour for equipment i (move/hr)
= time of service of berth (hr/day)
= number of berth
= time handing ratio for equipment i (%)
= utilization rate of equipment,
= maintenance & breakdown time (%)
� � = ��
��
+ �� + ��
di = average distance traveled per move of equipment i
vi = velocity of equipment i
tDi = time delayed
tTTi = time of transfer
--(5)
167
Jonathan Yong Chung Ee et al. / Jurnal Teknologi (Sciences & Engineering) 69:7 (2014), 163–171
Where, B, B12, B13 are coefficients of the ARIMA model (0,1,1)
(0,1,1)12.
3.2 Additive Holt-Winters Model
The "h-step ahead forecast" for the multiplicative Holts-Winter's
equation is a combination of an estimate of trend level (ℓ ),
growth rate (b ) and seasonal factor (sn )as below:
Ǭ
+ℎ
=ℓ +ℎ +
(h = 1,2,3,…) --(6)
+ℎ−
Each estimate factor formula of ℓ , b and sn can be
expressed as
ℓ =
b =
sn =
ℓ −ℓ
−
y −ℓ
−
−
+
+
−
−
+
−∝ ℓ
−
−
+
−
--(7)
--(8)
--(9)
−
Where α,, and are smoothing constants between 0.02< α, ,
< 0.9. And L is the number of seasons (e.g. L=4, for quarterly
data). (reference for constant range)
3.4 Accuracy Model Diagnostics
The issue of measuring the accuracy of forecasts from different
methods has been central of attention. Error is described in a
similar fashion as forecasting by one-step ahead forecast error,
which is simply et = Yt - Ýt, regardless of how the forecast was
produced. Therefore, the forecast h-step-ahead forecast error
iset+h= Yt+h - Ýt+h.
Common practice uses the basic scale-dependent measures
are based on absolute error or squared errors. Comparing mean
absolute error (MAE) and mean squared error (MAE), it is easier
to understand and compute the accuracy on the same scale. On the
other hand, percentage error have advantage of being scale
independent and is frequently used to compare performance
between different data sets. Below are the empirical expression of
the errors:
Mean Square Error
��� = ∑
3.3 Seasonal ARIMA Model
Example of an ARIMA(1,1,1)(1,1,1)s model (without a constant)
for s-lag data and can be written as:
( − ∑∅
=
).( − ∑Φ
=
).
−
.
−
�
= ( − ∑�
=
)
− ∑Θ
Mean Percentage Error
=
Basic steps to fitting ARIMA model to the forecast data can
be divided into a simple 5-step-procedure. First, plot the data
accordingly to its axis which may reveal some features that
indicates the pattern of seasonal and stationary. Second, choose to
transform the data by performing the natural logarithm function to
minimize the standard deviation. Then, the crucial step of
identifying the order (p,d,q) and (P,D,Q), if seasonal ARIMA is
used, must be done with care. The data plot may aide in
identifying the differencing order, d (beware of overdifferencing).
While inspection of the autocorrelation (ACF) and partial
autocorrelation function can help identify the AR order and MA
order. Fourth, estimation for the model parameter can be
performed using Yule-Walker equation or any time series
software such as SAS, Minitab and Statistica. However, the
maximum likelihood and method of least squares must be
observed. Lastly, the residual diagnosis must be done by
reviewing non-residuals' ACF and PACF, histogram that indicate
Gaussian white noise, else iteration has to be done by estimating
another set of orders (p,d,q).
Since the "ARIMA (0,1,1)(0,1,1) 12" is used, which denotes
a zero order autocorrelation, 1st order difference, 1st order
moving average, zero order seasonal autocorrelation, 1st order
seasonal difference, 1st order seasonal moving average. The
seasonal analysis period is 12, which is a monthly interval in a
year's period. Therefore, ARIMA(0,1,1)x(0,1,1)12 can be
expressed as:
.
−
=
−�
−Θ
+� Θ
−̂
---(12)
=
−̂
---(13)
� �= ∑
--(10)
−
Mean Absolute Error
� �= ∑
=
--(11)
Mean Absolute Percentage Error
� �= ∑
=
=
�� −�̂�
|�� −�̂� |
��
��
---(14)
---(15)
However, the most favored measure is the Mean Absolute
Percentage Error (MAPE) proposed by Makridakis (1993) [44].
4.0 RESULTS
4.1 Holts-Winter Result
Parameter grid search is performed by iterating for the least error
for the parameter chosen for the set of data. The above Table 2
shows that at 406th iteration, the mean absolute error(MAE), sum
of squared error (SSE), mean squared error (MSE) are the least
with other error indicators at relative low, hence, the parameter
alpha, delta and gamma is chosen. Then, Holts-Winter method is
performed using "Minitab".
Table 2 Iteration of parameter
168
Jonathan Yong Chung Ee et al. / Jurnal Teknologi (Sciences & Engineering) 69:7 (2014), 163–171
Forec as ts ; Model:(0,1,1)(0,1,1) Seas onal lag: 12
Input: TEU Throughput (ln) (-D1)(-D12)
Start of Year: 1995
End of Year: 2017
Winters' Method Plot for TEU Holts-Winter
Multiplicative Method
1750000
Variable
Actual
Fits
Forecasts
95.0% PI
1500000
TEU Holts-Winter
1250000
Smoothing Constants
Alpha (level)
0.6
Gamma (trend)
0.1
Delta (seasonal)
0.1
1000000
750000
Accuracy Measures
MAPE
5
MAD
23298
MSD
1061553046
500000
250000
1800000
1800000
1600000
1600000
1400000
1400000
1200000
1200000
1000000
1000000
800000
800000
600000
600000
400000
400000
200000
200000
0
95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17
19 19 19 19 19 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20
Index
Figure 2 TEU throughput of Holt-Winters forecast
0
0
Obs erv ed
The forecast (Figure 2) predicts a strong increase of
throughput in the coming years, based on unknown external
factors but only based on the information from the data set.
Residual Plots for TEU Holts-Winter
Normal Probability Plot
Residual
Percent
100000
90
50
10
Forec as t
± 90.0000%
Figure 4 TEU throughput of SARIMA forecast
Result (Figure 4) also shows a steady increasing pattern for
the TEU throughput forecast for 2012 to 2016. Below, Table 3
shows the parameter for seasonal and non-seasonal moving
average coefficient, also the indication of errors ensuring best fit
of model.
Versus Fits
99.9
99
0
1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017
0
-100000
1
Table 3 Parameter iteration results of least error
0.1
-100000
0
Residual
100000
200000
Histogram
100000
36
Residual
Frequency
800000
Versus Order
48
24
12
0
400000
600000
Fitted Value
0
-100000
-135000
-90000
-45000
0
45000
90000
Residual
95 96 97 98 99 00 01 02 03 04 05 06 07 08 09 10 11 12 13 14 15 16 17
19 19 19 19 19 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20
Observation Order
Figure 3 Normal probability, histogram and goodness of fit of results'
residual
The probability plot and histogram of residual-frequent
(Figure 3) indicates a healthy set of skewed and normally
distributed pattern for the set of data. The residuals versus fits
graph indicates nonconstant variance which spreads unevenly
across the fitted values. Residual versus order also fluctuates
around zero except the middle -end section fans, which is still
acceptable.
Following are the probability plot and histogram of residualfrequent (Figure 5, 6 & 7) which indicate a healthy set of skewed
and normally distributed pattern for the set of data. ACF and
PACF of the data residual (Figure 8 & 9) are also shown,
indicating acceptable trend, with only one spike at lag 3 for both
functions.
Normal Probability Plot: TEU Throughput
ARIMA (0,1,1)(0,1,1) residuals;
3
2
Data is analyzed by checking autocorrelation (ACF) and partial
autocorrelation (PACF) before carrying out ARIMA order
assignment. After analysis, function of natural logarithm and
difference in lag 1 and lag12 is required to ensure desirable ACF
and PACF pattern. Then, SARIMA (0,1,1)(0,1,1) model is
performed using "Statistica".
Expected Normal Value
4.2 Seasonal ARIMA (0,1,1)(0,1,1)12
1
0
-1
-2
-3
-0.30
-0.25
-0.20
-0.15
-0.10
-0.05
0.00
0.05
0.10
0.15
Value
Figure 5 Normal probability plot of SARIMA residual
0.20
169
Jonathan Yong Chung Ee et al. / Jurnal Teknologi (Sciences & Engineering) 69:7 (2014), 163–171
pattern of forecasting by both models having only slightdifference
in the maximum throughput forecast in August (Table 4). The
maximum throughput will be the targeted service throughput for
the fulfillment of equipment capacity.
Histogram; variable: TEU Throughput
ARIMA (0,1,1)(0,1,1) residuals;
Expected Normal
80
70
Partial Autocorrelation Function
VAR7 : ARIMA (0,1,1)(0,1,1) residuals;
(Standard errors assume AR order of k-1)
60
No of obs
50
40
30
20
10
0
-0.40
-0.30
-0.35
-0.20
-0.25
-0.10
-0.15
0.00
-0.05
0.10
0.05
0.20
0.15
0.30
0.25
0.35
Upper Boundaries (x<=boundary)
Figure 6 Histogram of SARIMA residual
Plot of variable: TEU Throughput
ln(x),D(1),D(12);
0.5
0.5
0.4
0.4
0.3
0.3
0.2
0.2
Lag
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
Corr. S.E.
-.109 .0724
+.046 .0724
+.293 .0724
-.056 .0724
-.080 .0724
+.087 .0724
+.008 .0724
+.098 .0724
+.036 .0724
-.057 .0724
+.142 .0724
-.070 .0724
+.004 .0724
-.037 .0724
+.060 .0724
-.101 .0724
+.032 .0724
+.068 .0724
-.045 .0724
+.102 .0724
-.031 .0724
+.085 .0724
+.152 .0724
-.035 .0724
+.040 .0724
+.008 .0724
-.071 .0724
-.018 .0724
+.049 .0724
-.038 .0724
-.021 .0724
+.018 .0724
-.019 .0724
+.033 .0724
-.075 .0724
+.141 .0724
+.062 .0724
-.051 .0724
+.032 .0724
+.133 .0724
-.034 .0724
+.030 .0724
+.134 .0724
-.010 .0724
+.016 .0724
-.091 .0724
+.107 .0724
-.055 .0724
+.034 .0724
+.038 .0724
0
-1.0
-0.5
0.0
0.5
1.0
Conf. Limit
Figure 9 Partial autocorrelation function for residual
TEU Throughput Forecast
9E5
0.1
0.0
0.0
-0.1
-0.1
-0.2
-0.2
-0.3
-0.3
-0.4
-0.4
VAR7
0.1
8E5
7E5
6E5
-0.5
-0.5
0
1996
1998
2000
2002
2004
2006
2008
2010
2012
5E5
4E5
Case Numbers
3E5
Figure 7 SARIMA residual plot
2E5
Autocorrelation Function
VAR7
Lag
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
Corr. S.E.
-.109 .0718
+.057 .0716
+.278 .0714
-.111 .0712
-.016 .0710
+.155
-.074 .0708
.0707
+.099 .0705
+.080 .0703
-.079 .0701
+.189 .0699
-.073
-.009 .0697
.0695
+.060 .0693
+.003 .0691
-.105 .0689
+.109 .0687
+.015 .0685
-.067
+.157 .0683
.0681
-.046 .0679
+.064 .0677
+.172 .0675
-.060 .0673
+.072
+.113 .0671
.0669
-.162 .0667
+.084 .0665
+.084 .0663
-.137 .0661
+.113 .0657
.0659
+.027
-.067 .0655
+.100 .0653
-.076 .0651
+.086 .0648
+.088 .0646
-.102
+.081 .0644
.0642
+.170 .0640
-.148 .0638
+.133 .0636
+.191 .0634
-.110
+.132 .0631
.0629
+.027 .0627
+.037 .0625
+.007 .0623
+.074 .0621
-.001 .0618
0
-1.0
1E5
1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 2017
: ARIMA (0,1,1)(0,1,1) residuals;
(Standard errors are white-noise estimates)
Q
2.30
2.94
18.10
20.52
20.57
25.35
26.44
28.41
29.72
30.99
38.33
39.42
39.44
40.18
40.18
42.50
45.02
45.07
46.03
51.36
51.81
52.71
59.18
59.97
61.11
63.98
69.88
71.49
73.10
77.40
80.37
80.54
81.57
83.91
85.29
87.04
88.91
91.42
93.00
100.1
105.4
109.8
118.9
122.0
126.3
126.5
126.9
126.9
128.3
128.3
0
-0.5
0.0
0.5
1.0
p
.1293
.2305
.0004
.0004
.0010
.0003
.0004
.0004
.0005
.0006
.0001
.0001
.0002
.0002
.0004
.0003
.0002
.0004
.0005
.0001
.0002
.0003
.0001
.0001
.0001
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
.0000
TEU Throughput
SARIMA Forecast
HW Forecast
Figure 10 Comparative TEU throughput forecast (HW vs SARIMA)
Table 4 TEU Forecast at Planning Time Horizon (2016)
Conf. Limit
Figure 8 Autocorrelation function for residual
For comparative Analysis of Holt-Winters and SARIMA
Forecast, both forecast are plotted on the same chart (Figure 10)
for seasonal trend observation. The results shows a closely fitted
Time
SARIMA Forecast
Holt-Winters Forecast
Jan-16
Feb-16
Mar-16
Apr-16
May-16
Jun-16
Jul-16
Aug-16
Sep-16
Oct-16
Nov-16
Dec-16
729111
645650
703509
740580
785465
770994
804231
844711
818067
848105
803020
753527
741943
643221
703077
738655
799407
784006
819898
857469
826333
845243
803711
746766
170
Jonathan Yong Chung Ee et al. / Jurnal Teknologi (Sciences & Engineering) 69:7 (2014), 163–171
4.3 Results of Empirical and 'Ratio' Method for Equipment
Estimation
Table 5 Estimation of equipment planning
Method/Time Series
Model
STS Crane
RTG
Prime
Mover
2011 (Actual)
74
97
933
2011 (Empirical)
69
153
306
Empirical – HW (2016PTH Forecast)
80
179
358
Empirical – SARIMA
(2016-PTH Forecast)
81
181
363
Ratio – HW (2016-PTH
Forecast)
80
160
400
Ratio – SARIMA (2016PTH Forecast)
81
162
405
Using the formulas and information on the equipment-crane
ratio, the numbers of required equipment estimation at planning
time horizon is tabulated in Table 5.
5.0 DISCUSSION
Though both the forecasting model proofed to be giving close
similarity in results, we note the superiority of the SARIMA model
for its flexibility of transforming and eliminating spikes in ACF
and PACF analysis. Since the set of TEU throughput history data is
a steady increase, both the forecasting model is considered as
having same prediction capability, though differing in concept.
Noting that the actual number of RTG and prime mover far
exceeds the empirical method, the factor in play here is the
equipment dynamics adopted by the specific port. The green and
sustainable trend of [45] equipment combination are mainly STS
crane with the support of RTGs and Prime Mover, which produce
lesser CO2 emission as compared to those which substitute RTGs
for forklifts, side-loaders , reachstackers, etc. Substituting 56
RTGs, the unnamed port compensated with 10 rail-mounted
gantries (RMG), 127 forklifts, 147 top-stackers and 37 sidestackers. However, possible varying consideration of parameter
assumption [46] has slight influence over the estimation of
equipment, nonetheless, the major reference for equipment
planning is the requirement of the STS crane.
The empirical method for equipment estimation has provided
a means to account for the conventionally ratio-determined
equipment profile. The 'check and balance' shows only a small
acceptable margin difference. Results also prove that the use of
more RTG can reduce the use of smaller equipment such as
forklifts, stackers and prime-mover, which are major contributors
of environmental waster. With the technology of electrification of
RTG should all the more drive port authorities to the use of 'stateof-the-art' technology for environmental purposes.
Since the TEU throughput forecast yields only a small margin
of 12758 TEU difference, the effect on equipment empirical
estimation has but little influence with differing 1 STS crane; 2
RTGs and 5 prime movers. At planning time horizon, the ratio
method of estimation yields about 10% less in deviation from the
empirical method. Hence, empirical method may provide for
redundancy that could be utilized in case of massive equipment
breakdown.
Since, the specific port has huge amount of prime-movers, the
planning ahead is to merely procure RTGs, or even eRTGs (fullyelectric or hybrid) to facilitate operation whilst slowly scraping
small equipments such as forklifts and top-stacker, etc as they wear
off. Not only will this reduce environmental waste but also greatly
reduce the operators of the many equipments reduced.
6.0 CONCLUSION
Though Holt-Winter and SARIMA (0,1,1)(0,1,1)12 methods yield
close results of forecasting, they are still univariate methods only
considering trends in the data set assuming no known factors
influencing it. Future equipment estimation should incorporate
multivariate models evaluating factors such as GDP, import-export
trend, population growth, immigration, inflation, etc [32].
For future consideration of equipment planning, it will be
interesting to incorporate a green and sustainable planning system
to perform equipment profile estimation with integration with other
elements of the port planning such as its hinterland, container park
area, container freight area, berthing zone and other terminal area,
etc [13].
Last but not least, an in-depth investigation on yard layout
area and equipment profile dynamics and operation parameters
should be evaluated properly or even a standard value for specific
type of port should be proposed. The effects on equipment costing
will be significant and will be of great concern to stakeholders in
the planning process.
Acknowledgement
We are grateful for the precious input from both supervisors and
colleagues from the specific field for contributing practicing data
and figures.
References
[1]
Steenken D., S. Vob, R. Stahlbock. 2004. Container Terminal Operation
and Operations Research: A Classification and Literature Review.
Spectrum. 26: 3–49.
[2] Young-Tae Chang , Jose Tongzon , Meifeng Luo & Paul Tae-Woo Lee.
2012. Estimation of Optimal Handling Capacity of a Container Port: An
Economic Approach, Transport Reviews. A Transnational
Transdisciplinary
Journal.
32(2):
241–258.
DOI:
10.1080/01441647.2011.644346
[3] Prakash Gaur. 2005. Port Planning as a Strategic Tool: A Typology,
Master Thesis of Science in Transport and Maritime Economics. Institute
of Transport and Maritime Management Antwerp University of Antwerp.
[4] Jasmine, L. S. L., Eddy, V. D. V. 2012. Green Port Strategy for
Sustainable Growth and Development. Nanyang Technological
University & University of Antwerp, Unknown Publisher.
[5] Joan C. Rijsenbrij, Armin Wieschemann. 2011. Sustainable Container
Terminals: A Design Approach. Handbook of Terminal Planning,
Operations Research/Computer Science Interfaces Series. 49: 61–82.
[6] Nima Sharif Mohseni. 2011. Master Thesis Project: Developing a Tool
for Designing a Container Terminal Yard, Delft University of
Technology & Royal Haskoning.
[7] Frankel, E. G. 1987. Port Planning and Development. Wiley-Interscience
Publication. ISBN 0-471-83708-3.
[8] Loke, K. B., Kader, A. S. A. and Zamani, M. 2012. Optimising
Expansion Control On Quay Crane Development In Container Terminal,
International Conference on Marine Technology Kuala Terengganu,
Malaysia, 20-22 October 2012, UMT/FMSM/MARTEC/MMP-17.
[9] MOMAF. 1998. Optimal Port Capacity Study on National Ports
(Contract Research Report Written in Korean) (Seoul: The Korea
Maritime Institute).
[10] Brown, R. G. 1963. Smoothing, Forecasting and Prediction of Discrete
Time Series. Prentice Hall, Englewood Cliffs, New Jersey
171
Jonathan Yong Chung Ee et al. / Jurnal Teknologi (Sciences & Engineering) 69:7 (2014), 163–171
[11] Holt, C. C. 2004. Forecasting Seasonals and Trends by Exponentially
Weighted Moving Averages. International Journal of Forecasting. 20:
5–10.
[12] Box, G. E. P., & Jenkins, G. M. 1976. Time Series Analysis: Forecasting
and Control. Hoboken USA: John Wily & son, Inc.
[13] UNCTAD. 1985. Port Development–A Handbook for Planners in
Developing Countries. United Nations, New York. 227.
[14] Kuroda, K., Takebayashi, M., Tsuji, T. and Muto, M. 1999. Influence of
Port Management Policy on the Behavior of Container Liner Ship
Company and Shippers. Journal of EASTS. 3(2): 127–138.
[15] Chu, C. Y. and W. C. Huang. 2005. Determining Container Terminal
Capacity on the Basis of an Adopted Yard Handling System. Transport
Reviews. 25(2): 181–199.
[16] Dally, H. K. 1983. Container Handling and Transport—A Manual of
Current Practice. London: Cargo Systems IIR Publications Ltd.
[17] Dharmalingam, K. 1987. Design of Storage Facilities for Containers—A
Case Study of Port Louis Harbour, Mauritius, Ports and Harbors,
September. 27–31.
[18] Itsuro, W. 2001. Container Terminal Planning—A Theoretical Approach.
Leatherhead, UK: World Cargo Publishing.
[19] Dekker Sander &Verhaeghe, R. J. 2008. Development of a Strategy for
Port Expansion: An Optimal Control Approach. Maritime Economic and
Logistics. 10: 258–274.
[20] Saman Kader. 1997. Cost Modelling for Inland Waterway Transport
Systems. Doctor of Philosophy in Marine Engineering. Liverpool John
Moores University. 276.
[21] Silver, E., D. Pyke, and R. Peterson, 1998. Inventory Management and
Production Planning and Scheduling. Wiley, USA.
[22] ZdenkaZenzerović, Edna Mrnjavac (Unknown). Modelling of Port
Container Terminal Using The Queuing Theory. Dept. of Maritime
Studies, University of Rijeka.
[23] Nevins, M. R., Macal, C. M., Love, R. J. and Bragen, M. J. 1998.
Simulation, Animation, and Visualization of Seaport Operations.
Simulation. 71(2): 96–106.
[24] Gambradella, L. M., Rizzoli, A. E. and Zaffalon, M. 1998. Simulation
and Planning of Intermodal Container Terminal. Simulation. 71(2): 107–
116.
[25] Yun, W. Y. and Choi, Y. S. 1999. A Simulation Model for ContainerTerminal Operation Analysis Using an Object Oriented Approach.
International Journal of Production Economics. 59(1–3): 221–230.
[26] Tahar, R. M. and Hussain, K. 2000. Simulation and Analysis for the
Kelang Container Terminal operations. Logistics Information
Management. 13(1): 14–20.
[27] Legato, P. and Mazza, R. M. 2001. Berth Planning and Resources
Optimization at a Container Terminal via Discrete Event Simulation.
European Journal of Operational Research. 133(3): 537–547.
[28] Nam, K.-C., Kwak, K.-S. and Yu, M.-S. 2002. Simulation Study of
Container Terminal Performance. Journal of Waterway, Port, Coastal,
and Ocean Engineering. 128(3): 126–132.
[29] Rijsenbrij, J. and A. Wieschemann. 2011. Sustainable Container
Terminals: A Design Approach. Handbook of Terminal Planning. J. W.
Böse, Springer New York. 49: 61–82.
[30] Chatterjee, S. and A. S. Hadi. 2012. Regression Analysis by Example. 5th
ed. New York: John Wiley & Sons.
[31] Duan X. et al. 2012. Application of Attribute Theory for Container
Throughput Forecast, IEEE International Conference on Granular
Computing 2012. DOI: 978-1-4673-2311.
[32] Li, X. and Xu, S. 2011. A Study on Port Container Throughput Prediction
Based on Optimal Combined Forecasting Model in Shanghai Port.
ICCTP 2011. 3894–3905.
[33] Rowe, G. 2007. A Guide to Delphi. Foresight: The International Journal
of Applied Forecasting. 8: 11–16.
[34] Xie, G. et al. 2013. Hybrid Approaches based on LSSVR Model for
Container Throughput Forecasting: A Comparative Study. Applied Soft
Computing. 13(5): 2232–2241.
[35] Chen, S.-H. and J.-N. Chen. 2010. Forecasting Container Throughputs at
Ports Using Genetic Programming. Expert Systems with Applications.
37(3): 2054–2058.
[36] Chou, C.-C., et al. 2008. A Modified Regression Model for Forecasting
the Volumes of Taiwan’s Import Containers. Mathematical and
Computer Modelling. 47(9–10): 797–807.
[37] Seabrooke, W. et al. 2003. Forecasting Cargo Growth and Regional Role
of the Port of Hong Kong. Cities. 20(1): 51–64.
[38] Peng, W.-Y. and C.-W. Chu. 2009. A Comparison of Univariate Methods
for Forecasting Container Throughput Volumes. Mathematical and
Computer Modelling. 50(7–8): 1045–1057.
[39] Gosasang, V. et al. 2011. A Comparison of Traditional and Neural
Networks Forecasting Techniques for Container Throughput at Bangkok
Port. The Asian Journal of Shipping and Logistics. 27(3): 463–482.
[40] Liu, C.-I., et al. 2004. Automated Guided Vehicle System for Two
Container Yard Layouts. Transportation Research Part C: Emerging
Technologies. 12(5): 349–368.
[41] Gosasang Veerachai, Watcharavee Chandraprakaikul, Supaporn
Kiattisin. 2010. An Application of Neural Networks for Forecasting
Container Throughput at Bangkok Port, Proceedings of the World
Congress on Engineering 2010 Vol I, WCE 2010, June 30 - July 2, 2010,
London, U.K.
[42] Winters, P. R. 1960. Forecasting Sales by Exponentially Weighted
Moving Averages. Management Science. 6: 324–342.
[43] Roland Fried, Ann Cathrice George. 2011. Exponential and Holt-Winters
Smoothing. International Encyclopedia of Statistical Science. DOI
10.1007/978-3-642-04898-2, Springer-Verlag Berlin Heidelberg.
[44] Makridakis, S., C. Chatfield, M. Hibon, M. Lawrence, T. Mills, K. Ord &
L. F. Simmons. 1993. The M2-competition: A real-time judgmentally
based forecasting study. International Journal of Forecasting. 9: 5–22.
[45] Sander Dekker. 2010. Sustainable Port Development In The
Netherlands–Framework for a Comprehensive Approach Applied To
Amsterdam Port. 12th Wctr Lisbon, Portugal. Strategic Consultant
Transport & Economics, Grontmij Nederland Bv.
[46] Park Nam-Kyu, Branislav Dragović. 2009. A Study of Container
Terminal Planning, Faculty of Mechanical Engineering, Belgrade. FME
Transactions. 37: 203–209.
[47] Zamani Ahmad. 2006. The Application of Fuzzy Expert System to
Preliminary Development Planning of Medium Size Container Terminal.
Doctor of Engineering in Engineering Business Management. Universiti
Teknologi Malaysia. 355.