\

(H
ij*
)
a5
6
*
a
ij
ijk
H
H


.

\

7
*
a
ij
ijk
D
D


.

\

(Y
ij
)
a8
(2.1)
Where Q
ijk
: Traffic flow between zone i to j via mode k;
P
i
, P
j
: Populations of i and j;
C
ij*
: Least cost of travel between i and j;
C
ijk
: Cost via mode k;
H
ij
: Shortest travel time between i and j;
H
ijk
: Travel time via mode k;
D
ij*
: Departure frequency of the most frequent mode;
D
ijk
: Departure frequency of mode k;
Y
ij
: Weighted average incomes of i and j;
a
o
,.., a
8
: Calibration parameter.
The transportationplanning model is primarily in estimating expected travel demands for
the purposes of design future transportation infrastructure system (Hellinga, 1994).
However, it has least two disadvantages that make it difficult for the estimation the OD
matrix for the operation and evaluation of the traffic management system. First, it requires
knowledge of studied zonal characteristics, which need knowing in previous years. These
data are costly, sometimes unavailable to obtain and can not reflect the detail of trip
making required for traffic management. Second, it is generally capturing the aggregate
changes in trip making behavior reflected by changes in zonal characteristics. It is
Zone 2 characteristics
Zone 1 characteristics
Origin
Destination
6
impossible to get the daily or hourly changes that are a part of the dynamic OD traffic
changes. Moreover, in fourstep models, the errors at one step are likely to magnify at
other steps because the inputs to one step are the outputs of another.
2.1.3 OD Demand Estimation from Direct Sampling
This approach is to directly measure the OD in the way of observing and analysis some
random data. From that, draw to the entire population.
LightOn Studies
Figure 2.3 Use LightOn Study to Estimate OD Demands
(Adopted from MsShane et al, 1990)
This method is use in case of only two origin points and maximum of two or three
destination points, generally within to 1 mile of each other (McShane et al, 1990). In
concept, Light On study is quite simple. Drivers on one arrival leg are asked to turn on
their headlights. At the exit pints, vehicles are counted and classified as lights on or
light off to determine their origin. As a practical matter, drivers on the entry leg with the
lowest volume are generally asked to turn their lights. Light On method is only used in
simple network and thus, impossible in applying urban transportation networks.
LicensePlate survey
Figure 2.4 Use License Plate Technique to Estimate OD Demands
(Adopted from Hellinga, 1994)
Generally, license plate technique is popularly used in direct sampling approach. Observers
record the license plate of the traffic passing through the observation location, which is
Leg A
Leg B
Leg C
Leg D
Zone 2
Zone 1
Observation
Location
Observation
Location 1
Observation
Location 2
Matching License Plates at
all Observation Locations
254
621
215
458
784
155
265
251
516
458
254
625
LightOn
sign
LightOff
sign
Count locations
7
located at numbers of zones in the network system. After that, license plate numbers
observed at different locations are used to match vehicles origin and destination.
In practice, because of time constraints, generally only a part of license plate numbers is
recorded. Due to this reason, the error occurring is not only in the sampling technique but
also incorrect vehicle identification.
This method is costly, time consuming in both observation and analysis and usually
reflecting only over peak periods. Even this approach can give the results better than
transportation planning process, it is still irrelevant for construction OD demand matrix for
the operation and evaluation of traffic management system.
Interview Methods
Interview methods are used to estimate OD demand matrix in a limited area. Some
common used interview study methods are roadside interviews, mailback questionnaires
and employee and major activity center interviews. These methods are quite simple.
However, there always have some bias; response rates are low and time consuming. So
they are difficult to construct OD demand matrix for the operation and evaluation of traffic
management system.
2.1.4 OD Demand Estimation from Traffic Counts
Figure 2.5 Use Link Count Technique to Estimate OD Demands
(Adopted from Hellinga, 1994)
The last category is the estimation OD demand matrix based on observed link traffic
counts and the one is used in this research. Difference from other categories above, by
using this method, the unknown OD demand would be reflected directly based on link
counts. The advantage of using link counts as the basics is that it can reflect the dynamic
temporal and spatial behavior of the demand. With rapid development of ITS, more and
more CBD have surveillance system in road network. Typically in the form of detector, it
is able to provide traffic flow continuously in real time. Frequently, these data are very
useful for operation and estimation of traffic management system.
However, due to the temporal nature of traffic flow and measurement error, observed link
flow are rarely consistent. Thus, there is no feasible solution that gives exactly the same
observed link flows. Furthermore, some approaches can not ensure that the estimated
Zone 1
Observation
Location
Zone 2
Link flow
measurement
Period of time
8
demand will not be nonnegative as the practice. Thus, when the solution is obtained, it can
not guarantee that this solution is optimal (Hellinga, 1994).
Another disadvantage of this approach is that when time series are considers, the traffic
flow is the function of time and level of demand. So in case of under congestion condition,
due to the movement of shockwaves, the data collected sometimes does not reflect the true
demands.
2.1.5 Most Likely Trip Matrix Estimation from Link Counts
Information Minimizing Approach
Information minimizing approach has been developed by Van Zuylen (1978) using
Brillouin measure (1962).
The information contained in a set of N observations where the state k has been observed
n
k
times is defied as:
I =



.

\

k
k
k
n
k
n
q
N
k
!
ln (2.1)
N : Number of observation;
K : Particular state;
n
k
: Number of times state k are observed.
The number of observed volumes between i and j by using link a can be expressed as
probability from i to j using link a multiplied by the total demand from i to j
n
ij
a
= P
ij
a
T
ij
(2.2)
P
ij
a
: Probability that demand from i to j by using link a;
T
ij
: Total demand from i to j.
Following Van Zuylen and Willumsen (1980), the probability that count on link a
associated with origin i and destination j can be describe as
q
ij
a
=
ij
a
ij ij
a
ij ij
P t
P t
) (
(2.3)
t
ij
: Prior number of trip between i and j;
q
ij
a
: Probability that count on link a associated with origin i and destination j;
P
ij
a
: Probability that demand from i to j by using link a.
Let denominator of equation (2.3) equals to S
a
S
a
=
ij
a
ij ij
P t ) ( (2.4)
9
Using Stirlings approximation lnX! = Xln(X) X , it is possible to have:
I 


.

\

a ij a ij
a ij
a
ij ij
V t
S T
P T
'
ln (2.5)
The objective is to determine T
ij
that maximize the information content of the estimated
demand (Hellinga, 1994), meaning that maximize I. Thus
Maximize I 


.

\

a ij a ij
a ij
a
ij ij
V t
S T
P T
'
ln (2.6)
Subject to V
a
=
ij
a
ij ij
P T ) ( (2.7)
V
a
: Total number of observed vehicle count on link a (vph);
t
ij
: Prior number of trip between i and j;
P
ij
a
: Probability that demand from i to j by using link a;
T
ij
: Total demand from i to j.
Entropy Maximizing Approach
Van Zuylen and Willumsen (1980) defied entropy as the measure of negative information
W = I 


.

\



.

\

k k
k
k
N q
n
n ln (2.8)
W : Measuring of entropy;
I : Measuring of information;
N : Number of information;
k : A particular state;
n
k
: Number of times state k are observed;
q
k
: The probability that state k occurs.
For applying maximizing approach, state k is considered to be the origindestination pair
that a trip utilizing. The number of times that state k are observed is equal to the demand
between i and j: n
k
= T
ij
(2.9)
T =
ij
ij
T = N =
ij
k
n (2.10)
Where T: Total number of trip;
T
ij
: Demand between origin i and destination j ;
N: Number of observation;
k: A particular state.
n
k
: Number of times state k outcomes are observed.
The probability that a trip is associated with a particular ij pair can be expressed as
q
ij
=
ij
ij
ij
t
t
(2.11)
where q
ij
: The probability that a trip is associated with ij pair;
10
t
ij
: Prior demand from origin i to destination j.
Substituting equation (2.9), (2.10), (2.11) into equation (2.8):
W = I 


.

\



.

\

ij ij
ij
ij
T q
T
T ln (2.12)
 T T
q
T
T
ij ij
ij
ij
ln ln +


.

\



.

\

(2.13)
The objective is to maximize equation (2.13) that satisfied the observed link count
constraints. Assuming that total number of trip is to be know. Thus:
Maximize W 


.

\



.

\

ij ij
ij
ij
T q
T
T ln (2.12)
subject to V
a
=
ij
a
ij ij
P T ) ( a (2.15)
Least Squared Error Approach
Least squared error approach (Hellinga, 1994) based on the minimizing of the sum square
differences from observed link flow and estimated link flow:
E =
2
) ' (
a
a
a
V V
(2.16)
The estimated flow on link a equals to the sum of all demand from origin i to destination j
is multiplied by the probability that demand between i and j using link a
V
a
=
ij
a
ij ij
P T ) ( (2.17)
Thus the partial derivative of V
a
with respect to unknown demand is equal to the
proportion of that demand contributing to the flow on link a
a
ij
ij
a
P
T
V
=
(2.18)
where V
a :
Estimated flow on link a(vph);
V
a
: Observed flow on link a(vph);
P
ij
a
: Probability that demand from i to j by using link a;
T
ij
: Demand between origin i and destination j (vph).
Substituting equation (2.17) into (2.16)
E =
2
'


.

\

a ij
a
a
ij ij
V P T (2.19)
11
The objective is minimizing total square differences from observed and estimated link
flow. Thus, minimize E. due to E is a convex function, it is possible to partial derivative
equation (2.19) with respect to each unknown demand and set each result equal to zero
ij
T
E
=


.

\



.

\

a
a
ij
ij
a
a
ij ij
P V P T ' 2 = 0, ij (2.20)
or


.

\



.

\

a
a
ij
ij
a
a
ij ij
P V P T ' = 0, ij (2.21)
Least Relative Error Approach
Developed by Hellinga in 1994, the Least relative error model is based on the minimizing
the total relative link flow differences:
E =
2
'
ln


.

\



.

\

a a
a
V
V
(2.22)
where E : Total relative link flow error;
V
a
: Estimated flow on link a (vph);
V
a
: Observed flow on link a (vph).
Since E is convex fuction, its minimum can be obtained by taking partial derivative of
object function with respect to each unknown demand and setting the each result equal to
zero
ij
T
E





.

\



.

\



.

\

a ij
a
a
a
a
T
V
V
V
V '
ln
'
ln 2 = 0 (2.23)
ij
T
E


.

\



.

\

a ij
a
a
a
T
V
V
V
'
ln 2 = 0 (2.24)
Moreover, the estimated flow on link a equals to the sum of all demand from origin i to
destination j is multiplied by the probability that demand between i and j using link a
V
a
=
ij
a
ij ij
P T ) ( (2.25)
V
a
: Estimated flow on link a (vph);
V
a
: Observed flow on link a (vph);
T
ij
: Demand between origin i and destination j;
P
ij
a
: Probability that demand from i to j by using link a;
Substituting equation (2.25) into equation (2.24)
12
ij
T
E
=
a
a
ij
V P
a a
a
V
V
/
'


.

\

=1 (2.26)
subject to V
a
=
ij
a
ij ij
P T ) ( (2.27)
2.1.6. Summary
A general review of the literature expressed some approaches to estimate OD matrix from
traffic count. Those solutions have advantages and disadvantages as well. No one shows
that clearly better than others. Information minimizing approach has several desired
characteristics. The model is insensitive to the number of trips represented by the seed
demand (Hellinga, 1994). When the seed demand feasible, the solution is that the estimated
traffic demand approximates the seed demand. In case of seed demand is not feasible, the
objective function is minimized when the estimated demand modifies the prior matrix.
However, this model also has a weakness. The model used simplified Stirlings
approximation x! x
x
e
x
but actually approximation is x! 2 x
x+1/2
e
x
(Feller, 1966)
and the difference both approximations is not well defied.
The Entropy maximizing approach is similar to the Information minimizing approach.
Thus the model makes full use of the information contained in the observed counts.
However, it also has some weaknesses in using Stirlings approximation and assuming
total number of trip to be a constant.
The Least Squared Error approach and Least Relative Error approach are based on the
assumption that the estimated demand should minimize the difference between estimated
traffic flows and observed traffic flows. It has been shown that those formulations are
sensitive to be the relative link flow error, regardless of whether this error results from
underestimation or an overestimation of the flow (Hellinga, 1994). The Least Squared
Error does not ensure that the demand is nonnegative. This model uses modified Jacobi
method but the relaxation factor ( ) is not well discussion. Unlike Least Squared error
approach, Least relative error does not lead to negative demand estimations. Unfortunately,
these nonlinear constrains can not be explicitly solved for the unknown OD demand and
hence requires an iterative solution and no mathematical proof is given that the algorithm
converges (Hellinga, 1994).
Several advantages and disadvantages of most likely approaches have been identified. Base
on these characteristics, Least square error approach are utilized and improved. Chapter 3
represents the mathematical formulation and iterative solution algorithm for a link count
based on Least Squared Error Model (LSE).
2.2 Traffic Assignment
The result of trip assignment is traffic volumes that will occur at each link on the network
within the study area. The most used traffic assignment model is equilibrium assignment.
In the equilibrium assignment, there are usually several equally good paths through the
network for each original destination pair (Yue et al, 2000). Wardrops first principle
(1952) express that drivers try to choice the paths to minimize their own travel times in the
network. This is also called principle of user optimization or principle of user equilibrium.
13
Wardrops second principle (1952) express that traffic is assigned to minimize total travel
time of all drivers in the whole network. This is called principle of system optimization.
Besides equilibrium assignment, the concept of nonequilibrium assignment, in which the
path for each origin destination is the shortest path, are also necessary because the used
equilibrium assignment procedures usually use a nonequilibrium assignment as a sub
component (Yu, 1994). Nonequilibrium models include allornothing assignment, multi
path assignment, diversion method and capacityrestrained assignment.
2.2.1 Allornothing Assignment
Allornothing assignment is based on the theory that a motorist or transit user will select
the quickest route between any OD pair. In other word, the traveler will always select the
route that represents minimum travel time. Thus, to determine which route that will be
chosen, it is necessary to find the shortest route from the zone of origin to all other
destination zones.
The general mathematical algorithm that describes the process is to select paths that
minimize the expression
ij
ij ij
T V (2.28)
where V
ij
: Volume on link ij;
T
ij
: Travel time on link ij;
i,j : Adjacent nodes.
2.2.2 Multipath Assignment
Allornothing assignment assumes that all drivers travelling between a pair of zones select
the same path. In real world, when all trips are assigned to a shortest path, this path will
increase travel time. Therefore, it is more realistic to assign trips between these paths
according to some proportion. Irwin et al (1962) suggested the following inverse
proportion function to calculate the percentage of trips between i and j will use route r and
this model is called Multipath Assignment
=
x
x
ij
r
ij
r
ij
t
t
P
1
1
) (
) (
(2.29)
where
r
ij
P : Percentage of trips to be assigned to route r;
r
ij
t : Travel time of route r from origin i to destination j;
x : The number of routes from i to j.
2.2.3 Diversion Method
The most widely used diversion procedure in the 1960s was that available in the series of
traffic planning computer programs of the Federal Highway Administration (Dickey,
1983). This method involves two travel times versus volume relationships to refer to the
14
allocation of a trip interchange to two possible routes in a designated proportion that
depends on some specified criterion.
Travel Time Ratio
Figure 2.6 Time via Arterial Highway Time via Quickest Alternative Route
(Adopted from Dickey, 1983)
The most widely used diversion technique in early studies was developed by Campbell.
Figure 2.6 is based on relative travel times between the new facility and the quickest
alternative route.
2.2.4 CapacityRestrained Assignment
Capacityrestrained assignment assumes that the link travel time is a function of the
assigned link volumes. In a capacityrestrained assignment, the trips are initially assigned
to shortest paths between a pair of zone that are calculated based on free flow link travel
times. Then the link travel times are recalculated based on a flow travel time function that
reflects the nature of the relationship between link flow and link travel time. New shortest
paths are subsequently calculated and the traffic volumes are reassigned to new shortest
paths.
The Bureau of Public Roads (BPR) function has been widely used to calculate the link
travel times in capacityrestrained assignment. BPR function is expressed mathematically
as
t =
(
(

.

\

+
C
V
t
o
1 (2.30)
where t : The travel time of a given link at flow of V;
t
o
: Free flow travel time of the link;
V : The link flow in vph;
C : The link capacity in vph;
, : Dimensionless coefficients.
0
20
40
60
80
100
0.3 0.5 0.7 0.9 1.1 1.3 1.5 1.7
50
50% Usage
Major roads
Freeways
Z
o
n
e
t
o
z
o
n
e
U
s
a
g
e
o
f
A
r
t
e
r
i
a
l
H
i
g
h
w
y
a
s
(
%
)
1
15
The magnitudes of coefficients and are a function of the characteristics of specific link
that is being modeled, where a typical value of is 0.15 and of is 4 (Vythoulkas, 1990)
To improve the accuracy of coefficients and , Drew had developed an iterative solution
called Drews technique. This method represents the relationship between travel time ratio
and volume to capacity ratio as
mr
o
mr
ij
t
t
=
mr
ij
mr
ij
Q
V
(2.31)
where V
mr
ij
: Volume of trip from i to j using m over r;
Q
mr
ij
: Capacity in vph;
t
mr
ij
: Travel time from i to j using m over r for a volume V
ij
mr
;
t
mr
o
: Free flow travel time from i to j using m over r.
Drews technique is expressed mathematically as
t
mr
ij
= t
mr
o
(
(
mr
ij
mr
ij
r
mr
ij
mr
ij
Q V
k Q V
/ 1
) 1 )( / [( 1
(2.32)
where k
r
: A level of service factor which varies for different types of highway
facility;
0 k
f
where f is freeway or expressway;
k
a
1 where f is arterial route;
k
r
= 0 for a perfect highway facility;
2.2.5 Summary
All or nothing assignment is simplest method to conduct traffic assignment. It is commonly
used and adequately demonstrates the basic principle involved. However, it is unrealistic
due to some basic assumptions. In real situation, the travel time will be increased when all
trips are assigned to a shortest path. Thus, it is useful when applied for complex road
network and common used for computer application. Multipath assignment is more
realistic to assign trips between these paths according to some proportion. The Diversion
method is also the simple method but this method is applied in only the simple network
(usually two road alternatives) and the curve in the figure should be modified for different
situation. Capacityrestrained assignment is the most accurate method in traffic
assignment. It is iterative procedure and represents the relationship between travel times
and capacities. However, this method is very time consuming and impossible to apply for
large networks. Considering the time consuming and the accuracy, Multipath assignment is
conducted in this research.
16
CHAPTER 3
METHODOLOGY
3.1 Conceptual Framework
In traffic engineering, most of problems are solved usually based on the aggregated link
measurements such as average speed, average delay and average travel time. Linklevel
data require trip level data, such as trip origin, trip destination, and trip departure time.
However, triplevel data are usually unavailable and costly to obtain directly from
surveying data. This research examines the problem of estimating unknown triplevel data
(origin  destination traffic demands) from readily available linklevel data. The actual
traffic flow in road network in Sukhumvit area is simulated then improved by changing or
adding some traffic policies. The concept to conduct this research is divided into seven
major states as follow:
Assessing the problems of existing conditions of the traffic system in Sukhumvit
area and collecting traffic data including road geometry, travel time, signal control
system and traffic volume;
Identifying link use probability;
Developing OD demand matrix from link traffic volume based on the theory of
least squared errors;
Estimating the accuracy of Least Square Model;
Constructing distributions of traffic flow and generating random variables;
Simulating the traffic condition by using TRAFNETSIM package;
Based on traffic assignment, evaluating some traffic policies;
The simple diagram of methodology is described as below:
Figure 3.1 Simple Diagram of Methodology
The detail of methodology is described as flowchart as Figure 3.2
Data
collection
Develop OD
demand matrix
from link counts
Simulate traffics
TRAFNETSIM
package
Evaluate some
traffic policies
17
3.2 Data Collection
The data used in this study includes link geometry, link traffic volume, link travel time and
link traffic signal control. These information is collected from secondary data (BMA,
2000) and surveying (January 2002). The time conducted is peak hour afternoon from
5pm to 6pm. At the time considered, large amount of vehicles using this network causes
congestion commonly.
3.2.1 Link Geometry
Link geometry includes length of link, number of lanes, lane width, lane channelization,
lane configuration, length of turning pocket and gradient. In this research, lengths of links
were measured from Bangkok map, other link characteristics are conducted from
surveying. This data is used for simulation the study network.
Data collection
 Network representation
 Road geometry
 Link traffic flows
 Link travel times
 Traffic control systems
OD matrix estimation
 Least square error estimation
 Link use probabilities
 Solution algorithm
 Pascal Source Code Program
Simulation
Simulation package
TRAFNETSIM
Traffic assignment
Change traffic policies
Evaluate traffic policies
and infrastructure
development
Secondary data and
Observation
Link flow distribution
Figure 3.2 Flowchart of Methodology
18
3.2.2 Link Traffic Flow
The traffic flow at each link is measured from secondary data (Data Collection at
Intersection, BMA, 2000). The traffic flows were found with mixture of vehicle types.
Trucks and buses are larger than passenger cars, thus occupy more road space. The number
of large and medium vehicles was converted as equivalent number of passenger cars in
order to achieve common unit. The impact of buses and trucks were treated through the
using of passenger car equivalents (PCE). The Table 3.1 below is shown PCE values
currently used in Thailand, which is adopted from Mathetharan, 1997.
Table 3.1 Passenger Car Equivalent (PCE) for Medium and Large Vehicles
Vehicle PCE
Motorcycle 0.25
Passenger car 1.00
Taxi 4wheel
TukTuk
1.00
0.75
Bus
Light
Medium
Heavy
1.25
1.50
2.00
Truck
4wheel
6wheel
10wheel
Articulated
1.75
1.75
2.00
3.00
Source: Mathetharan, 1997
Thus, each link traffic flows are computed approximately by setting equal to average value
of two approaches connected this link.
3.2.3 Link Travel Time
In order to conduct link travel time measures, a traditional floating car study was carried
out to collect data describing trip duration along each link. Due to budget constrain, the
motorcycle was used and run with the same speed as a typical car in traffic flow. In case of
some links have more than one lane, the motorcycle was rode in the midlane or inner lane.
The travel time will present the total time required for a vehicle to pass the link. This time
will include time spent travelling and time lost in a queue due to congestion or a traffic
signal. Thus, the congestion will effect on travel time. The link travel time will be used to
estimate link use probability later.
3.2.4 Traffic Signal Control
At the peak periods, in Sukhumvit area, usually polices control traffic signal system by
themselves. Base on their experience, they will permit vehicles at some approaches to go
and others are stopped. In this situation, one model should be developed to estimate their
behaviors. However, due to time constraint, traffic signal systems in this research are
chosen as pretime signal control.
19
3.3 OriginDestination Traffic Demand Estimates from Link Traffic Flow
3.3.1 Definition of OD Estimation Based on Link Traffic Flow
Traditionally, in the transportation planning, a static OD matrix is assumed that traffic
volumes are the input of the traffic assignment procedure. To determine the link flow
traffics, the OD demand is computed. This process requires the route choice behaviors and
the link flows at the time consideration. Several methods have been used to estimate OD
traffic demand, regardless of the method, the result will be represented by the matrix which
describes the proportion of demand departing from origin i to destination j by using link a.
V
a
=
ij
a
ij ij
P T (3.1)
Where i : Origin zone number;
j : Destination zone number;
a : Link number;
V
a
: Flow observed arriving on link a at time o;
a
ij
P : Proportion of demand departing origin i to destination j that will arrive on
link a;
T
ij
: Demand departing from origin i to destination j.
For time varying OD demands (dynamic OD demands), the matrix includes two more
disturbances, the departing time of the demands at the origin and the time at which
observed demand is on link a . The need of dynamic demand estimation can be established
when OD demand exhibit significant temporal and spatial variations (Hellinga, 1994). At
that situation, the assumptions which are made in the estimation of static demand are
increasingly violated as the time is more disaggregated. These violations lead to
inaccuracies of static demand estimation. However, the estimated dynamic demands also
need more required data, which is costly and time consuming. In this study, the data is
observed in separated one hour. Considering the timing and budget required, the static OD
demand estimation is prevailing and will be conducted.
Equation (3.1) contains three values: link flows, OD demands and link use probabilities.
In the OD estimation procedure, traffic flow at links is observed directly. The link use
proportions are achieved from link travel times. The OD demands are unknown and are
the objectives of the process.
3.3.2 Identifying Link Use Probability
Irwin et al (1962) suggested the following inverseproportion function to calculate the
percentage of trips between i and j will use route r and this model is called Multipath
Assignment
=
x
x
ij
r
ij
r
ij
t
t
P
1
1
) (
) (
(3.2)
The probability that demand between i and j will use link a is equal to sum of all
probabilities of trips between i and j will use route r which includes link a:
20
=
x
x
ij
a
ij
P P (3.3)
where
r
ij
P : Probability that demand between i and j will use route r;
a
ij
P : Probability that demand between i and j will use link a;
r
ij
t : The travel time of route r from origin i to destination j;
x : The number of routes from i to j.
Algorithm of Multipath assignment are described as follow
Step 1: Identify all routes which users can go from i to j including directed routes
(using only one link) and undirected routes (using more than one link);
Step 2: Calculate the travel time of each route which users can go from i to j;
r
ij
t =
a
a
t
Step 3: Calculate the probability of routes which users can go from i to j;
=
x
x
ij
r
ij
r
ij
t
t
P
1
1
) (
) (
Step 4: Calculate probability that demand between i and j will use link a
=
x
x
ij
a
ij
P P
Step 5: Continue with other i,j pairs until all pairs are checked.
Where i,j : Origin, Destination;
r
ij
P : Probability that demand between i and j will use link a;
r
ij
t : The travel time of route r from origin i to destination j;
a
t : The travel time of link a;
x : The number of routes from i to j.
a : The number of links of specified route r;
3.3.3 Least Squared OD Estimation Model (Hellinga, 1994)
The Least Squared model applied for OD demand estimation, which is adopted from
Hellinga, 1994 is one of the simplest methods for OD demand estimation based on link
traffic volumes. However, the author did not explain how to determine the relaxation factor
(), which make the model converges. In this research, this value are estimated and proved
mathematically.
21
3.3.3.1 Model Formulation
The least squared error (LSE) model described is based on the estimated OD demand,
which minimize the squared link flow differences.
Link flow error function as the sum of the squared difference between estimated and
observed link flow is derived as follow:
E =
a
a a
V V
2
) ' ( (3.4)
The objective is to identify the solutions that minimize E. Because E is a convex function
and differentiable, it is possible to get minimize E by computing the partial derivative of E
with respect to each unknown OD demand (T ij ) and setting each equation equal to zero:
ij
T
E


.

\

a ij
a
a a
T
V
V V ) ' ( 2 = 0 (3.5)
Moreover, the estimated link flow on link a is the sum of the demand between origin i to
destination j with the proportion of that demand using link a
V
a
=
ij
a
ij ij
P T (3.6)
The partial of derivative of V
a
with respected to a unknown demand T
ij
is the proportion of
that demand using link a.
ij
a
T
V
= P
ij
a
(3.7)
Because values of the demands are the positive numbers so T
ij
0,
ij
(3.8)
Substitute equation (3.6), (3.7) and inequality (3.8) into equation (3.5) the result is:
ij
T
E
=


.

\

a
a
ij
ij
a
a
ij ij
P V P T ' ) ( = 0,
ij
(3.9)
T
ij
0,
ij
(3.10)
where: a : Link identifier;
V
a
: Estimated flow on link a (vph);
V'
a
: Observed flow on link a (vph);
a
ij
P : Probability that demand between i and j will use link a;
T
ij
: Demand between origin i and destination j (vph).
3.3.3.2 Development of a Solution Algorithm
The solution is based on iterative process and modified Jacobi method. By using a
relaxation technique (Hornbeck, 1975), after an iteration is completed, these computed
values are modified by factor (1) prior to use them in the next iteration. For the general
22
OD estimation case, the modified demand T
ij
estimation for the current is provided as
follow:
T
ij
l+1
= T
ij
l
 ) (
ij
ij
C
n
(3.11)
Where n
ij
=
a
a
ij
P
2
) ( (3.12)
C
ij
=
a
a
ij a a
P V V ) ' ( (3.13)
T
ij
l
: The modified demand from origin i to destination j at iterative l;
V
a
: Observed flow on link a (vph);
V
a
: Estimated flow on link a (vph);
P
ij
a
: Probability that demand between i and j will link a;
: Relaxation factor (0 < < 1);
l : Iterative number.
Algorithm steps:
Step 1 : Set the iteration counter to zero : l = 0;
Step 2 : Use the prior matrix as the initial estimate of the unknown demand matrix
for iteration zero: T
ij
= t
ij
,
ij
Step 3 : Assign the current estimate of the demand to the network using known link
use probabilities V
a
=
ij
a
ij ij
P T ,
ij
Step 4 : Compute the sum of the difference between estimated and observed link
flows for all links associated with each OD pair C
ij
= ( )
a
a
ij a a
P V V ' ;
Step 5 : Increment the iteration counter l = l+1;
Step 6 : Compute new estimates of all ODs
) (
1
ij
ij
l
ij
l
ij
C
n
T T
=
+
n
ij
=
a
a
ij
P
Step 7 : Incorporate nonnegativity constrains If T
1 + l
ij
<0 then T
1 + l
ij
=0;
Step 8 : Check stopping criterion, if not meet go to step 2;
3.3.3.3 Algorithm Stopping Criteria
Since the algorithm above is iterative, stopping criteria must be defined to determine when
the iterations are sufficient to carry out the result. An appropriate measure of the marginal
reduction in link flow error is obtained by computing the difference between the relative
link flow error of the previous iteration with the current iteration.
23
N E E / =
(3.14)
' /
= V E E
n
(3.15)
5 1
10 ) ( ) (
< =
l n l n n
E E E (3.16)
where
N : Number of links;
'
ij
a
ij ij
P T (3.17)
ij
C =


.

\

a
a
ij a
a
ij
a ij
a
ij ij
P V P P T '
= const P P T
a
ij
a ij
a
ij ij
+


.

\

(3.18)
1 + l
ij
T = ) (
ij
ij
l
ij
C
n
T
=
ij
l
ij
n
T
(
(
+


.

\

const P P T
a
ij
a ij
a
ij
l
ij
=
l
ij
a
a
ij
ij
T P
n


.

\

2
) ( 1
ij
n
(
(
(
+



.

\

<> <>
const P P T
a
ij
a
j k i m
k m
a
mk
l
mk
,
,
(3.19)
According to Hornbeck (1975) convergence of the GaussSeidel iterative method cannot
be guaranteed unless
<>
=
>
n
j i
j
ij ii
C C
1
(3.20)
Let (3. 19) into (3. 20)
2
) ( 1 (
a
a
ij
ij
P
n
>
<> <>
ij a
a
ij
j k i m
k m
a
mk
ij
P P
n
) . (
,
,
(3.21)
24
Due to both left hand side and right hand side of (3. 22) are positive, it can be removed
absolute value on both side of the equation, thus
2
) ( 1
a
a
ij
ij
P
n
>
<> <>
ij a
a
ij
j k i m
k m
a
mk
ij
P P
n
.
,
,
2
2
2
,
,
) (
) ( 1
) ( . 1
<
>



.

\

+ >
<> <>
a ij
a
ij
ij
a ij
a
ij
ij
ij a a
a
ij
a
ij
j k i m
k m
a
mk
ij
P
n
P
n
P P P
n
Thus, in order to make LSE model converge, relaxation factor () should satisfy equation
(3. 23). Moreover,
a ij
a
ij
P
2
) ( N (Number of link) (3.24)
Thus,
N
n Min
ij
) (
(3.25)
Finally, the relaxation factor is determined as follow
+ Assign equal to minimum of 0.5 and n
ij
;
+ Divided by the number of link in that network.
(3.23)
25
No
No
Yes
E
n
= (E
n
)
l1
 (E
n
)
l
<10
5
l <>1
No
Yes
l := 0
T
ij
a
:= t
ij
=
ij
a
ij ij a
P T V
=
a
a
ij a a ij
P V V C ) ' (
l:=l+1
ij
ij
l
ij
l
ij
C
n
T T
=
+1
=
a
a
ij ij
P n
2
) (
:= min(, N
ij
)
:= /N
If T
ij
l+1
< 0 T
ij
l+1
= 0
V
a
: Estimated link flow at link a; V
a
: Observed link flow at link a;
t
i
: The prior demand from origin i to destination j; T
ij
: Estimated demand from origin i to destination j;
P
ij
a
: Probability of demand between i to j using link a ; V : Average observed link flow;
: Relaxation factor; L : The current iteration number.
N : Number of link; E : The least squared objective function.
Input
V
a
, P
ij
a
, t
ij
Output T
ij
Figure 3.3 Flowchart of OD Demand Estimation
26
3.3.4 Link Flow Distribution
For traffic flow in urban network is too complex, variety and traffic flow data taken from
secondary data (BMA, 2000) and observation is only one sample of population. Therefore
traffic flow distribution at each link is developed in order to generate random variables for
evaluating level of variety of traffic condition before and after changing traffic policies.
Due to lack of sufficient information, it is assumed following triangular density function
and each distribution is independent each other. The maximum value is the link capacity.
The mean is the value taken from secondary data. The minimum value based on condition
of each link and at the time considered. Then, ten cases for generating random variables
following triangular distribution at each link are obtain to estimate different OD demands.
3.3.4.1 Link Capacity
The capacity of a roadway link is defined as the maximum hourly rate at which persons or
vehicles can be reasonably expected to traverse a point or uniform section of a lane or
roadway during a given time period under prevailing roadway, traffic and control condition
(TRB, 1985). The transportation network analysis procedure uses hourlybased link
capacities due to the travel behavior of drivers are more accuracy.
Pignataro (1973) provided a widely used formula for obtaining freeway link capacities by
multiplying the level of service E volume (2000 passenger car per hour per lane) by
appropriate adjustment factor. Pignataros equation for obtaining freeway link capacities is
C = 2000NWTc (3. 26)
Where
C: capacity (mix vehicle per hour, total for one direction);
N: number of lanes (in one direction);
W: adjustment for lane width and lateral clearance;
Tc: truck factor at capacity.
However, link capacity is effected strongly by intersection. The vehicles still waiting for
service or decelerating at intersection will reduce the number of vehicles passing this link.
To express this reduction, the Rs = 0.45 will be multiplied into equation (3. 27).
C = 2000NWTcRs (3. 28)
The value of W is referred to Table 7.2 Principles of Highway Engineering and Traffic
Analysis. The value of Tc depends on the percentage of truck in traffic volume and less
than 1. The number of lanes is obtained from surveying.
3.3.4.2 Triangular Distribution
Figure 3.4 Triang(a,b,c) Density Function
X (Veh/h)
2/(ba)
a c
b
f(x)
27
Density
) )( (
) ( 2
a c a b
a x
if a x c
) )( (
) ( 2
c b a b
x b
if c < x b
Distribution 0 if x < a
) )( (
) (
2
a c a b
a x
if a x c
) )( (
) (
1
2
c b a b
x b
if c < x b
1 if b < 1
Mean
3
c b a + +
Variance
18
2 2 2
bc ac ab c b a + +
Mode c
3.3.4.3 Generating Random Variables
A simulation having any random aspects at all must involve sampling, or generating
random variables from one or more distributions (Averill et al, 1982). In this part,
assuming traffic flows are independent each other, random variables are generated with
triangular distribution in order to run the simulation model.
Averill et al (1982) had generated random variables of triangular as follow
Firstly, if X triang(0,1, (ca)/(ba)) then X = a + (ba)X triang(a,b,c), so it can be
restricted attention to triang(0,1,c) random variables, where 0 < c < 1. The distribution
function is easily inverted to obtain, for 0 u 1,
(cu)
1/2
1 [(1c)(1u)]
1/2
Therefore, it can be stated the following inversetransform algorithm for generating
X triang(0,1,c):
1. Generating U U(0,1);
2. If U c, set X = (cU)
1/2
and return. Otherwise, set X = 1 [(1c)(1U)]
1/2
and
return.
3.4 TRAFNETSIM: Traffic Simulation Software
Many Transportation System management (TSM) strategies affect the model and route
choice of tripmakers. In order to test the effect of TSM schemes on trip patterns, it is
necessary to analyze an area that contains a substantial portion of routes that the trip
makers may follow. This concept of a single integrated simulation system that can provide
f(x) =
F(x) =
F
1
(u) =
28
the user with flexibility and ease of use and that can optimize the efficiency of all
computations was conceived by the Federal Highway Administration (FHWA) in the mid
1970s. FHWA has since supported a series of projects to implement this design and to
develop the software for TRAF (TRAF users manual).
NETSIM, the model was integrated with TRAF simulation system in 1980 is one
component model in TRAF designed to represent traffic on a Urban street networks and at
microscopic level. Microscopic simulation models represent the traffic flow in term of the
individual vehicular movement. The concept is based on route choice behavior and car
following theory. The advantage of those models is that the trip users have full data in the
system network over time and space.
The TRAFNETSIM also includes a traffic assignment, which is designed to expand the
applicability of traffic simulation modeling to transportation planning. Traffic assignment
models can use for two purposes: to convert OD matrix into actual network and to evaluate
demand responses when changing policies. Thus, it is the key to estimate the traffic
policies in this research.
In the TRAFNETSIM, two optimization techniques are used in the equilibrium traffic
assignment model: the users optimal assignment and the systems optimal assignment.
The users optimal assignment has been reached when no driver can reduce his travel time
by choosing alternative routes. The systems optimal assignment is the minimum total cost
of the entire network. The algorithm that is used is allornothing traffic assignment.
Shortest path tree is constructed for each specified origin node to all other network nodes
by using labelcorrecting algorithm.
3.4.1 Evaluation of the Accuracy of LSE Model
To estimate the accuracy of LSE model, the comparisons between estimated and observed
link flows will be made. The objective is to estimate OD demand that closely replicate the
true demand. Since the true demand is unknown, it is impossible to make any direct
comparison between the estimated and true demand. However, it is possible to use traffic
assignment to compute the traffic flow at each link from determined OD matrix. Then,
measure the difference between estimated and observed traffic flow at each link. Applying
the OD demand based on LSE into TRAFNETSIM, then conducting traffic assignment,
the estimated flows are achieved as the output. Thus, the comparison between observed
and those estimated link flows are obtained.
3.4.2 TRAF NETSIM Parameter
Many parameters are important to generate the simulation data and to present the real field
effectively. The parameters should be calibrated by using the real traffic data which is
obtained by surveying.
Due to time constraint, the parameters are adopted from previous study (Surasak, 1997).
These parameters are reasonable values of Bangkok urban area and therefore can be
applied in this research. The TRAF NETSIM parameters used in this study to present the
traffic condition of Sukhumvit area are summarized as bellow:
29
Saturation flow rate 1800vph
Departure headway 2 sec.
Startup lost time 3.5 sec.
Free flow speed 40mph = 64.372kph
Traffic condition By using PCE to estimate OD demand, thus
 Passenger car 100%
 Bus 0%
 Truck 0%
Vehicles characteristics change only passenger cars length, the other as default.
Amber time 3 sec/phase
All red time 1 sec.
30
CHAPTER 4
DATA ANALYSIS
4.1 Network Representation
An aggregate representation of the study road network is covered by Phetchaburi Rd. from
North, Phrama IV from South, Ratchadamri Rd. from West and Ekamai from East.
Carrying large number of vehicles, this road system is most heavily traveled urban road
network and ranks among busiest in Bangkok. It attracts high volume of traffic as it is a
major road system for CBD and as it is the primary route for through traffic. However,
continuously increasing traffic demand has made this road system beyond its original
design capacity. It has caused into recurring congestion most of the time in a.m. and p.m.
business periods. Nonrecurring congestion is also common place throughout these peak
periods for both private vehicles and commercial automobiles.
The network was abstracted into a series of directed links and nodes. Nodes are shown on
Figure 4.3 and separated into internal nodes and external nodes described as bellow
External node: Intersection among the road segments inside and outside network,
including trip generations and trip attractions, and called as origindestination.;
Internal node: Intersections linked road segments inside network only.
External node 2,10,14 represent the on ramp and off ramp of Second state expressway. To
improve the accuracy of simplified network, node 6, 7, 8 are put to represent the complex
road (soi) system and shown as Figure 4.3. These nodes also include trip generations and
trip attractions and known as external nodes. Links are defined as the road segments
connected between two nodes. A summary of the number of links and nodes required to
present the network is provided in Table 4.1
Table 4.1 Number of Nodes and Links in Study Area
Type Number
Node Origin
Node Destination
Internal Node
Link
16
16
22
97
The GIS map of the existing Bangkok road network system is shown in Figure 4.1. A
defined study area and study network is shown Figure 4.2 and Figure 4.3.
31
Figure 4.1 The GIS map for the Bangkok road network system (not to scale)
4.2 Description of the Traffic Management System
The good transportation planning depends on the effectiveness of traffic management
system. Travel time, travel cost, congestion reduction, etc. belong to the traffic
management system. In the study area, the traffic management system includes contraflow
bus lane, signal control system, one way roads and reverse lanes.
The traffic signal utilized in the road network can be classified into two types: Pretime
control and Police control. Pretime control has fixed green time, amber time and red time.
It will automatically change the traffic pattern from green to red and vice versa at the
installed time. Police control means that polices control traffic volumes manually. This
method is easily seen onramp and offramp of Second stage expressway. Depending on
traffic condition at each approach, the polices give the priority to some approaches in the
way that permit passing through the intersection as well as stop vehicles from other
approaches. The more experienced polices are, the better traffic can be managed in
comparison with pretime control. However, it is not proved and out of scope of this
research.
In the study network, the exclusive bus lanes and contraflow bus lanes are used. By
reducing the travel time and increasing the accessibility of bus, it encourages people in
using public transport. However, due to the headway of bus is low (E.g. In Phloen Chit Rd.
contra flow bus lane, from 5pm to 6pm, only 38 buses used (BMA, 20
th
Dec. 2000)),
sometimes the bus land occupancy is free but land acquisition for other modes are
congested.
Sukhumvit
Area
32
On Phetchaburi Rd, based on the experiencing congestion due to lack of capacity to handle
heavy directional peak period traffic flow west to east, reversible lanes have been
implemented for better utilization of the available pavement. By using this technique, one
lane is operated in different directions at different times of day so as to provide more lanes
in the peak direction. However, it might seem that the use of reversible lanes could
possibly lead to an increase in some type of accidents (head on and sideswipe type
collisions) due to confusion which lane to be in, particularly at the changed lane.
In addition to traffic management system, one way street is implemented in Phloen chit
and other roads to reduce congestion, or just as a means of providing additional traffic
capacity and improve traffic safety. However, it may increase travel distance as well.
33
3
2
Figure 4.2. The Study Area
Legend
: Origin, Destination.
: Internal node.
: Link
Figure 4.3 The Study Network
29
21
17
18
19
20
30
31
22
23
24
25
26
28
27
29
32
33
34
35
36
37
38
15
14
16
13
12
7
4
9
10
8
11
1
6
5
34
4.3 OD demand estimation
Base on Least Square Error model, the program (Pascal source codes) has developed to
estimate OD demands. By using the iterative solution, after 1914 times, the marginal
reduction in link flow errors is lower than 10
5
. Then, the OD demands are obtained and
these values are used for next calculation.
Figure 4.4 Marginal Reduction in Link Flow Error
The indication of correlation between observed and those estimated link flows is estimated
by another program (Pascal source code) named P9.pas. The computed correlation
coefficient between the observed and estimated link flows is 0.5820. The high level of
correlation indicates that there is a strong association between the estimated and observed
link flows. However, these values must be considered with some caution because the
observed data are not all strictly independent. This program indicate that the LSE model
tends to overestimate observed flows at low flows (less than 2000vph) and underestimate
at high traffic flows (more than 3000vph). The result can be explained that due to traffic
flows can not be negative, errors on links with low flow tend to result from an over
estimation of the flows. The accuracy of estimated flows would reduce if observed traffic
flows are higher than 3000vph. The error happened due to inaccurate observation of travel
time at each link, which affected to probability of demand from origin to destination by
using a specified link, and the seed matrix. Due to the time and budget constraint, in this
research, seed matrix is chosen as uniform matrix.
The OD matrix then applies to TRAFNETSIM. Figure 4.5 shown the different between
observed and estimated traffic flow at each link.
0.00000
0.00005
0.00010
0.00015
0.00020
0.00025
0.00030
0.00035
0.00040
0
1
0
0
2
0
0
3
0
0
4
0
0
5
0
0
6
0
0
7
0
0
8
0
0
9
0
0
1
0
0
0
1
1
0
0
1
2
0
0
1
3
0
0
1
4
0
0
1
5
0
0
1
6
0
0
1
7
0
0
1
8
0
0
1
9
0
0
Iterations
M
a
g
i
n
a
l
r
e
d
u
c
t
i
o
n
i
n
l
i
n
k
f
l
o
w
e
r
r
o
r
35
Figure 4.5 Correlation between observed and those estimated link flows
Random variables are generated following triangular distribution at each link to represent
the variety of traffic flows. Ten OD demand matrixes are built based on those variables to
estimate the effectiveness after changing policies. The total origindestination traffic flows
in entire network of each case are described as bellow
Table 4.2 Total OriginDestination traffic flows
Case 1 2 3 4 5 6 7 8 9 10
Total OD 60728 57967 58466 60977 56380 57100 62281 58578 55777 60070
Then, the OD demand matrix has been used to input to TRAFNETSIM package. Traffic
assignment has been conducted.
4.4 Evaluation of Changing Policies
The change of policies are conducted by three approaches and described bellow
First approach: Oneway roads in the study network are changed to twoway
roads in the way of total number of lanes at each link are unchanged;
Second approach: New twoway roads are built to connect Sukhumvit Rd. with
Phetchaburi Rd. at node 6 and node 7. The characteristic of other roads are kept the
same existing condition;
Third approach: Phloen Chit Rd. from oneway road is changed to twoway road.
One new road is built to connect Sukhumvit Rd. and Phetchaburi Rd. at node 7.
In each approach, ten cases that generate random variables based on triangular distribution
is conducted. Three links (219) (Phloen Chit from Chitlom to Ratchadamri), (1721)
(Chitlom from Phetchaburi to Phloen Chit) and (194) (Phetchaburi from Phrom Phong to
Asoke) checked to determine how travel time of these links change after traffic policies
y = 1.0269x
R
2
= 0.5822
0
1000
2000
3000
4000
5000
6000
7000
8000
0 1000 2000 3000 4000 5000 6000 7000 8000
Observed flow (vph)
E
s
t
i
m
a
t
e
d
f
l
o
w
(
v
p
h
)
36
29
21
17 18
19
20
30
22
23
24
25
26
28
27
29
32
33
34
35
36
37
38
15
14
16
13
12
7
4
9
10
8
11
1
6
5
Figure 4.6 The Network after First Approach Implementation
change. The comparison of Total VehicleKilometer, Average speed and Average Travel
Time between before and after change policies would evaluation of the policies and
infrastructure development.
4.4.1 Change Roads from Oneway to Twoways
Phloen Chit Rd., Chitlom Rd., Withayu Rd., Phetchaburi Rd., Nana (soi 3), Asoke (soi 21),
Ratchadamri Rd. and Kluaynamthai are changed from oneway to twoways in the way of
keeping the same total number of lanes at each road (the changed links are represented in
red color). The network is shown on Figure 4.6.
Legend
: Origin, Destination.
: Internal node.
: Link
31
2 3
37
0
1
2
3
4
5
6
7
8
9
10
1 2 3 4 5 6 7 8 9 10
Case
A
v
e
r
a
g
e
T
r
a
v
e
l
T
i
m
e
(
M
i
n
/
K
m
)
Existing
1st
Figure 4.7 Average Travel Time of link (21 9) before and after
First Approach Implemetation
0
2
4
6
8
10
12
14
16
18
1 2 3 4 5 6 7 8 9 10
Case
A
v
e
r
a
g
e
T
r
a
v
e
l
T
i
m
e
(
M
i
n
/
K
m
)
Existing
1st
Figure 4.8 Average Travel Time of link (17 21) before and after
First Approach Implemetation
0
2
4
6
8
10
12
1 2 3 4 5 6 7 8 9 10
Case
A
v
e
r
a
g
e
T
r
a
v
e
l
T
i
m
e
(
M
i
n
/
K
m
)
Existing
1st
Figure 4.9 Average Travel Time of link (19 4) before and after
First Approach Implemetation
38
Figure 4.12 Comparision of Average Travel Time, before and after
First Approach Implemetation
0
1
2
3
4
5
1 2 3 4 5 6 7 8 9 10
Case
A
v
e
r
a
g
e
t
r
a
v
e
l
t
i
m
e
(
M
i
n
/
K
m
)
Existing
1st
Figure 4.10 Comparision of Total VehicleMile, before and after
First Approach Implemetation
0
10000
20000
30000
40000
50000
60000
70000
80000
1 2 3 4 5 6 7 8 9 10
Case
T
o
t
a
l
V
e
h

K
m
Existing
1st
Figure 4.11 Comparision of Average Speed, before and after
First Approach Implemetation
0
2
4
6
8
10
12
14
16
1 2 3 4 5 6 7 8 9 10
Case
A
v
e
r
a
g
e
s
p
e
e
d
(
K
P
H
)
Existing
1st
39
From Figure 4.7, it is shown that after changing, the travel time of link (21 9) is
increased to double. The reason is that this link attracts more route users in both directions.
The signal control system at node 21 and 9 are more complex due to adding more
directions. The conflict points increasing as well as turning movements more difficulty are
contributing to the result. It is the same explanation for link (1721), travel time is also
increased approximately twice. However, considering link (194) from Figure 4.9, travel
time before and after change is slightly different. It can be explained in way that this link is
quite far from changed links thus, the influence of changing policies is insignificant.
From Figure 4.10, it is clearly to see that after change from oneway to two way, the total
vehicle mile will be decreased. The total vehiclemile represents the product of total
number of vehicles and total mileage traveled in the network. This unit, thus, is the total
mileage traveled by all vehicles. Due to the total vehicle travelling in this network are
constant for each case, the reason is that route users can reduce the travel distance between
origin to destination. Traveler can choose alternative route since they can drive two
directions instead of one direction as before. For example, from Figure 4.7, vehicles
traveled from node 9 to node 10 can use link 30,31,32 directly instead of link 9, 1, 2, 3, 4,
12, 33. From origin 2 to destination 1, link 3,2,1 are used as an alternative of link 3, 2, 10,
30, 9. However, average speed is reduced and average travel time is increased as Figure
4.11 and Figure 4.12. It can be explained that in oneway roads, the conflict points at
intersections are less than in case of twoway roads. Converting relatively oneway roads
to twoway roads can greatly complexity the intersection operation and signal phasing. In
this network, the converting of roads told above will make problems of intersection
conflicts and delays significantly, especially at intersections Phloen Chit Ratchadamri,
Phloen Chit Chitlom, Phloen Chit Witthayu, Phloen Chit Nana. With oneway traffic
flow, turning movements are able to proceed unimpeded by opposing traffic. McShane et
al, 1990 cited of Homburger et al, 1982 reported a conversion of twoway to oneway
operation that resulted in
37% reduction in average trip time;
60% reduction in number of stop;
38% reduction in accidents.
The percentage of effectiveness after changing to twoway traffic are represented as follow
() : Decrease; (+) : Increase
Table 4.3 Percentage of Effectveness of Total VehKm,
Before and After First Approach Implementation
Case 1 2 3 4 5 6 7 8 9 10
Existing 68912 72016 72512 71822 71729 71166 74683 71302 79501 71116
2way 67464 63571 61709 62314 62961 63508 66890 62337 62659 62286
% effectiveness 2.10 11.73 14.90 13.24 12.22 10.76 10.44 12.57 21.19 12.42
The total VehicleKm will reduce at minimum 2.1%, maximum 21.19% and average
12.16% if the traffic policies are changed from oneway roads to twoway roads.
40
Table 4.4 Percentage of Effectveness of Average Speed (KPH),
Before and After First Approach Implementation
Case 1 2 3 4 5 6 7 8 9 10
Existing 13.45 13.45 13.78 13.53 13.76 14.07 13.29 13.84 14.19 13.68
2way 14.05 13.24 13.41 13.00 13.79 13.52 12.91 13.58 13.89 13.52
% effectiveness 4.43 1.56 2.69 3.92 0.23 3.89 2.91 1.86 2.15 1.18
Except from case 1
st
and case 5
th
, the average speed trends to reduce slightly in other cases.
The average effectiveness of Average Speed is reducing1.55%.
Table 4.5 Percentage of Effectveness of Average Travel Time (Min/Km),
Before and After First Approach Implementation
Case 1 2 3 4 5 6 7 8 9 10
Existing 4.46 4.46 4.36 4.43 4.36 4.26 4.52 4.33 4.23 4.39
2way 4.28 4.53 4.47 4.62 4.35 4.44 4.65 4.42 4.32 4.44
% effectiveness 4.18 1.53 2.71 4.21 0.14 4.08 2.89 2.01 2.21 1.13
The average travel times trend to increase slightly in most of cases. The average
effectiveness of Average Travel Time is increasing 1.65%.
After change most of one way streets to twoway streets in the study network, the total
vehiclemile will decrease average 12.16%. However, the Average Speed will decrease
slightly average 1.55%. The Average Travel Time will increase 1.65%. Before considering
this approach, the disadvantage about traffic safety and average speed, average travel time
should be taken into account. When this approach is implemented, it can cause more traffic
accidents due to increase number of conflict points, increase travel time and make traffic
congested at intersections as well.
41
3 2
4.4.2 Build New Twoway roads
Two new roads are built to connect Sukhumvit Rd. and Phetchaburi Rd. at node 6, 7. After
changed, the network is shown on Figure 4.14 (the new roads are represented in red color).
Legend
: Origin, Destination.
: Internal node.
: Link
From Figure 4.14 and Figure 4.15, after built more roads, travel time of link (21 9) and
link (17 21) are slightly reduced. It can be explained that route users would shift to new
routes so the traffic volumes at other routes reduce, thus travel time would reduce as well.
However, due to they are quite far from new roads, the effectiveness is low. Differently, in
Figure 4.16, the travel time of link (19 4) is increase twice after build new road. It can be
explained that after new roads are built, number of vehicles coming to node 40 is more
than that of coming out. It tends to transfer large amount of vehicles from Ploen Chit Rd.
and nearby to Phetchaburi Rd. Thus, it is better if new roads are built and connecting Ploen
Chit Rd. and Phetchaburi Rd., which would increase the transportation efficiency between
those both main roads.
The Total Vehicle Kilometer, Average Speed and Average Travel Time of entire network
are shown on Figure 4.17, 4.18 and 4.19
Figure 4.13 The network after Second Approach Implementation
39
40
29
21
17 18
19
20
30
31
22
23
24
25
26
28
27
29
32
33
34
35
36
37
38
15
14
16
13
12
7
4
9
10
8
11
1
6
5
42
Figure 4.14 Average Travel Time of link (21 9) before and after
Second Approach Implementation
0
1
2
3
4
5
6
1 2 3 4 5 6 7 8 9 10
Case
A
v
e
r
a
g
e
T
r
a
v
e
l
T
i
m
e
(
M
i
n
/
K
m
)
Existing
2nd
Figure 4.15 Average Travel Time of link (17 21) before and after
Second Approach Implementation
0
0.5
1
1.5
2
2.5
1 2 3 4 5 6 7 8 9 10
Case
A
v
e
r
a
g
e
T
r
a
v
e
l
T
i
m
e
(
M
i
n
/
K
m
)
Existing
2nd
Figure 4.16 Average Travel Time of link (19 4) before and after
Second Approach Implementation
0
5
10
15
20
25
1 2 3 4 5 6 7 8 9 10
Case
A
v
e
r
a
g
e
T
r
a
v
e
l
T
i
m
e
(
M
i
n
/
K
m
)
Existing
2nd
43
Figure 4.17 Comparision of Total vehiclemile, before and after
Second Approach Implementation
0
10000
20000
30000
40000
50000
60000
70000
80000
1 2 3 4 5 6 7 8 9 10
Case
T
o
t
a
l
V
e
h

K
m
Existing
2nd
Figure 4.18 Comparision of Average Speed, before and after
Second Approach Implementation
0
2
4
6
8
10
12
14
16
1 2 3 4 5 6 7 8 9 10
Case
A
v
e
r
a
g
e
s
p
e
e
d
(
K
P
H
)
Existing
2nd
Figure 4.19 Comparision of Average Travel Time, before and after
Second Approach Implementation
0
1
2
3
4
5
1 2 3 4 5 6 7 8 9 10
Case
A
v
e
r
a
g
e
t
r
a
v
e
l
t
i
m
e
(
M
i
n
/
K
m
)
Existing
2nd
44
Following results of TRAFNETSIM in Figure 4.17, Figure 4.18 and Figure 4.19, it is
realistic to see that, in most of cases, even if travel distance is increased, route users would
shift their routes to increase average speed and reduce travel time.
The percentage of effectiveness after changing to twoway traffic are represented as follow
() : Decrease; (+) : Increase
Table 4.6 Percentage of Effectveness of Total VehKm,
Before and After Second Approach Implementation
Case 1 2 3 4 5 6 7 8 9 10
Existing 68912 72016 72512 71822 71729 71166 74683 71302 79501 71116
New roads 68955 70826 71808 70537 66821 69363 72147 71068 65544 70801
% effectiveness 0.06 1.65 0.97 1.79 6.84 2.53 3.40 0.33 17.55 0.44
From Table 4.6, in most of cases, Total VehKm is reduced. The reason is that after second
approach is implemented, route users can shift their route to reduce travel distance and
travel time. In case 5
th
and 9
th
, when Total Origin  Destination traffic flows in the network
is low, the reducing in Total VehKm is very high because travelers can choose shorter
travel distance as well as reduce travel time. Differently, in other cases, some travelers
choose new road as other roads are congested. In this case, new roads are not option for
reducing travel distance but for reducing travel time. Thus, TotalKm is reduced slightly.
Except from case 1
st
, the Total VehicleMile will decrease at minimum 0.33%, maximum
17.55% and average 3.94% if the new roads are built as the Figure 4.13.
Table 4.7 Percentage of Effectveness of Average Speed (KPH),
Before and After Second Approach Implementation
Case 1 2 3 4 5 6 7 8 9 10
Existing 13.45 13.45 13.78 13.53 13.76 14.07 13.29 13.84 14.19 13.68
New roads 14.26 14.27 15.24 14.51 14.58 15.22 14.66 14.42 14.43 14.66
% effectiveness 6.02 6.10 10.59 7.24 5.96 8.17 10.31 4.19 1.69 7.16
The Average Speed is increased at minimum 1.69%, maximum 10.59% and 6.74%
average.
Table 4.8 Percentage of Effectveness of Average Travel Time (Min/Km),
Before and After Second Approach Implementation
Case 1 2 3 4 5 6 7 8 9 10
Existing 4.46 4.46 4.36 4.43 4.36 4.26 4.52 4.33 4.23 4.39
New roads 4.20 4.20 3.94 4.13 4.11 3.94 4.09 4.16 4.15 4.01
% effectiveness 5.93 5.83 9.63 6.77 5.73 7.51 9.51 3.93 1.89 8.66
The Average Travel Time is decreased at minimum 1.89%, maximum 9.63% and 6.54%
average.
Before this approach is implemented, the cost for construction, environmental impacts, and
the effectiveness of increase average speed, decrease travel time should take into account.
45
3
2
The next approach introduce the better solution, with decrease Total VehicleMile and
increase Average Speed and reduce Average Travel Time as well.
4.4.3 Build New Twoway Road accompany with Change Oneway to Twoway
Roads
A new twoway road is built to connect Sukhumvit Rd. and Phetchaburi Rd. at node 7.The
oneway Phloen Chit Rd. is changed to twoway road. The network after changed is shown
on Figure 4.20.
Legend
: Origin, Destination.
: Internal node.
: Link
Figure 4.20 The Network after Third Approach Implementation
From Figure 4.21 and Figure 4.22, Travel time of link 219 and link 1721 are increased
when changing to both direction instead of one direction. However, because only Ploen
Chit Rd. changed, the increase of travel time at this link is lower than that of first approach.
Moreover, route users can shift their routes from other roads to Ploen Chit Rd. The result is
that traffic flows at those roads reduced, thus travel time and travel distance would be
reduce as well.
The changing in travel time of link 194 is shown on Figure 4.23. Certainty, after this
new road is introduced, travelers would use this one to transport from Ploen Chit Rd. to
Phetchaburi Rd. Thus, travel time and travel distance of other roads nearby would reduce.
After changing policies, Total VehicleMile, Average Speed and Average Travel Time are
shifted as Figure 4.24, 4.25 and 4.26.
39
29
21
17 18
19
20
30
31
22
23
24
25
26
28
27
29
32
33
34
35
36
37
38
15
14
16
13
12
7
4
9
10
8
11
1
6
5
46
Figure 4.21 Average Travel Time of link (21 9) before and after
Third Approach Implementation
0
1
2
3
4
5
6
1 2 3 4 5 6 7 8 9 10
Case
A
v
e
r
a
g
e
T
r
a
v
e
l
T
i
m
e
(
M
i
n
/
K
m
)
Existing
3rd
Figure 4.22 Average Travel Time of link (17 21) before and after
Third Approach Implementation
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
1 2 3 4 5 6 7 8 9 10
Case
A
v
e
r
a
g
e
T
r
a
v
e
l
T
i
m
e
(
M
i
n
/
K
m
)
Existing
3rd
Figure 4.23 Average Travel Time of link (19 4) before and after
Third Approach Implementation
0
2
4
6
8
10
12
14
16
18
1 2 3 4 5 6 7 8 9 10
Case
A
v
e
r
a
g
e
T
r
a
v
e
l
T
i
m
e
(
M
i
n
/
K
m
)
Existing
3rd
47
Figure 4.24 Comparision of Total vehiclemile, before and after
Third Approach Implementation
0
10000
20000
30000
40000
50000
60000
70000
80000
1 2 3 4 5 6 7 8 9 10
Case
T
o
t
a
l
V
e
h

K
m
Existing
3rd
Figure 4.25 Comparision of Average speed, before and after
Third Approach Implementation
0
2
4
6
8
10
12
14
16
1 2 3 4 5 6 7 8 9 10
Case
A
v
e
r
a
g
e
s
p
e
e
d
(
K
P
H
)
Existing
3rd
Figure 4.26 Comparision of Average Travel Time, before and after
Third Approach Implementation
0
1
2
3
4
5
1 2 3 4 5 6 7 8 9 10
Case
A
v
e
r
a
g
e
t
r
a
v
e
l
t
i
m
e
(
M
i
n
/
K
m
)
Existing
3rd
48
The percentage of effectiveness after changing to twoway traffic are represented as follow
() : Decrease; (+) : Increase
Table 4.9 Percentage of Effectveness of Total VehKm,
Before and After Third Approach Implementation
Case 1 2 3 4 5 6 7 8 9 10
Existing 68912 72016 72512 71822 71729 71166 74683 71302 79501 71116
Mix 66567 69727 67877 66844 68209 68599 69943 67139 67139 63548
% effectiveness 3.40 3.18 6.39 6.93 4.91 3.61 6.35 5.84 15.55 10.64
The total vehiclemile would decrease at minimum 3.4%, maximum 15.55% and average
6.68% if the policies are as the Figure 4.20.
Table 4.10 Percentage of Effectveness of Average Speed (KPH),
Before and After Third Approach Implementation
Case 1 2 3 4 5 6 7 8 9 10
Existing 13.45 13.45 13.78 13.53 13.76 14.07 13.29 13.84 14.19 13.68
Mix 14.60 15.53 15.06 15.42 15.35 15.43 14.92 15.50 15.50 15.55
% effectiveness 8.49 15.43 9.35 13.91 11.58 9.73 12.23 11.98 9.18 13.65
The Average speed would increase at minimum 8.49%, maximum 15.43% and average
11.55%.
Table 4.11 Percentage of Effectveness of Average Travel Time (Min/Km),
Before and After Third Approach Implementation
Case 1 2 3 4 5 6 7 8 9 10
Existing 4.46 4.46 4.36 4.43 4.36 4.26 4.52 4.33 4.23 4.39
Mix 4.11 3.87 3.97 3.89 3.91 3.89 4.02 3.95 3.87 3.86
% effectiveness 7.94 13.37 8.84 12.20 10.27 8.75 11.00 8.75 8.38 12.04
The Average Travel Time would decrease at minimum 7.94%, maximum 12.20% and
average 10.15% if the third approach is implemented.
4.4.4 The Comparison of Three Approaches
Total vehicleMile, Average Speed and Average Travel Time are conducted to find what is
the best solution among three approaches. From Figure 4.27, Total Vehicle Mile of first
approach is lowest. It is meant that after changing to all twoway roads, travel distances
will be reduced. However, the Average Speed is lowest and Average Travel Time is
highest compared with other approaches. The reason is that after changing to twoway
roads, the conflict points at intersections will increase, turning movement is more difficult
so it takes more time, and congestion will be increased as well.
Second approach has highest Total VehicleMile. It can be explained that some route users
use new roads as shortest path for both travel distance and travel time. Besides, some
travelers shift their routes to decrease travel time even if increasing travel distance.
49
Figure 4.27 Comparision of Total VehKm of three approaches
0
10000
20000
30000
40000
50000
60000
70000
80000
1 2 3 4 5 6 7 8 9 10
Case
T
o
t
a
l
V
e
h

K
m
1st
2nd
3rd
Figure 4.28 Comparision of Average Speed of Three Approaches
0
2
4
6
8
10
12
14
16
1 2 3 4 5 6 7 8 9 10
Case
A
v
e
r
a
g
e
s
p
e
e
d
(
K
P
H
)
1st
2nd
3rd
Figure 4.29 Comparision of Average Travel Time of Three Approaches
0
1
2
3
4
5
1 2 3 4 5 6 7 8 9 10
Case
A
v
e
r
a
g
e
t
r
a
v
e
l
t
i
m
e
(
M
i
n
/
K
m
)
1st
2nd
3rd
50
The third approach is also the best approach. Compared to other approaches, the Total
vehiclemile is lower than existing condition, Average Speed is higest and Average Travel
Time is lowest.
The confident interval after ten cases conducted for Total VehicleMile, Average speed and
Average Travel Time is in Figure 4.30, 4.31 and 4.32.
The figure is shown that the lowest Average Speed of third approach is higher than that of
existing condition. The higest Average Travel Time of third approach is lower than that of
existing condition. It means that in all cases considered, after traffic policies are changed in
way of third approach, the traffic condition would be improved.
51
Figure 4.30 Interval of Total VehKm for each Approach
79501
67464
72147
69943
68912
65544
63548
74207
68846
66745
61709
64587
60000
64000
68000
72000
76000
80000
Existing 1st 2nd 3rd
T
o
t
a
l
V
e
h

K
m
Max
Min
Average
Figure 4.31 Interval of Average Speed for each Approach
14.19
13.89
15.24
15.55
13.29
12.91
14.26
14.60
13.74
13.40
14.75
15.07
12.00
13.00
14.00
15.00
16.00
Existing 1st 2nd 3rd
A
v
e
r
a
g
e
S
p
e
e
d
(
K
m
/
h
)
Max
Min
Average
Figure 4.32 Interval of Average Travel Time for each Approach
4.52
4.65
4.20
4.11
4.23
4.32
3.94
3.86
4.37
4.48
4.07
3.98
3.80
4.00
4.20
4.40
4.60
4.80
Existing 1st 2nd 3rd
A
v
e
r
a
g
e
T
r
a
v
e
l
t
i
m
e
(
M
i
n
/
K
m
)
Max
Min
Average
52
CHAPTER 5
CONCLUSIONS AND RECOMMENDATIONS
5.1 Conclusions
Traffic Management is often the most costeffective solution to transportation problems,
when all impacts are considered. Traffic Management can provide multiple benefits,
including reduced congestion, parking facility cost savings, accident cost savings,
consumer cost savings, pollution reduction, and more efficient land use. To estimate the
effectiveness of Traffic Management, one of the most analytical tools is traffic simulation,
which is possible to integrate several facilities and networks into one simulation model.
The main objective of the study is to determine better solution for traffic policies in
Sukhumvit area, Bangkok, Thailand. The area includes three main roads: Phetchaburi
Phloen Chit, Sukhumvit Phraram IV and covered from Ratchadamri Rd. to Sol 63
(Ekamai). Data  road geometry, travel time, traffic volume and traffic control system
were collected from field surveying on January 2002 and from secondary data: Bangkok
Metropolitan Administration, 2000. The details of this study are:
 Estimate OD demand based on link counts;
 Evaluate and estimate the effectiveness of some policies.
The major conclusions derived from the study and some recommendation are described as
bellow:
5.1.1 OD Demand Estimation
One program (Pascal Source Codes) is developed to identify the link use probabilities from
link travel time by using theory of multipath assignment. The second program (Pascal
Source Codes) based on the theory of Least Squared Error (LSE) is built to estimate the
OD demand according to traffic flow distribution and the knowledge of link use
probabilities above. The accuracy of the model is evaluated by the difference between the
observed and estimated link flows drawn by third program (Pascal Source Codes). The
computed correlation coefficient between the observed and estimated link flows is R =
0.5820. The research indicated that LSE model tends to overestimate observed flows at
low flows (less than 2000vph) and underestimate at high traffic flows (more than
3000vph). Moreover, the accuracy of estimated flows would reduce if observed traffic
flows are higher than 3000vph.
Since traffic flow in urban network is too complex, variety and data taken from secondary
data and observation is only one sample of population. Thus, traffic flow distribution is
developed at each link. Assume that flows at each link are followed triangular distributions
and they are independent. Thus, ten random variables following the triangular distribution
of each link are generated to reflect the variety of traffic flows. Finally, ten OD demands
are generated.
53
5.1.2 Effectiveness of Changing Policies
To estimate and improve traffic policies, three approaches for changing policies are
discussed in this study as follow
 First approach, Phloen Chit Rd., Chitlom Rd., Withayu Rd., Phetchaburi Rd, Nana
(Soi 3), Asoke (Soi 21), Ratchadamri Rd. are changed from oneway to twoway
and total number of lanes for both direction of these roads are unaltered.
Characteristics of other roads and traffic signal control systems are kept the same as
before. After conducted the first approach, The result showed that Total Vehicle
Kilometer reduces 12.16%. Average speed reduces about 1.55% and average travel
time increases 1.65% compared with existing condition.
 Second approach is that two new roads are built to connect Sukhumvit Rd. and
Phetchaburi Rd. Other policies are unchanged. The second approach represented
that Total vehiclemile would decrease 3.94%. Average speed would increase
average 6.74% and average travel time would reduce 6.54%.
 A new twoway road connected Sukhumvit Rd. and Phetchaburi Rd. and the one
way Phloen Chit Rd. changed to twoway road is the third approach. After
simulated, this approach indicated that Total Vehiclemile would reduce 6.68%
average. Average Speed would increase average 11.55% and Average Travel Time
would reduce 10.15%.
Finally, the research indicates out of three approaches, third approach is most
effectiveness. Conducting ten cases per each approach, the result showed that in average
travel time, the worse case of third approach (4.11 min/km) is lower than the best case of
existing condition (4.23 min/km). In average speed, the worse case of third approach (14.6
km/h) is higher than the best case of existing condition (14.9 km/h).
5.2 Related Topic for Further Studies
This study provided some ideas related to traffic policy improvement as well as
infrastructure development in Sukhumvit road network. Although several findings have
been found in this study, it is useful to conduct further research as follow:
In the research, benefits of changing policies and infrastructure development are
mentioned. However, it is better if further studies estimate costs, then compare between
benefits and costs;
Further researches should analyze the environmental impacts before and after changing
traffic policies and infrastructure development;
Link traffic volumes discussed in the research are assumed dependence on each other.
Thus, further study should consider the correlation among link traffic volumes;
54
In the research, the improvement of traffic condition is based on changing oneway
roads to twoway roads and constructing new roads. Further studies can estimate other
policies such as traffic control system at some critical intersections;
Related topic for further research should improve from Static OD Demand Estimation
to Dynamic OD Demand Estimation and the time consideration should be extended
from one hour to longer to represent the actual traffic condition more accuracy;
This result is obtained from ten cases running with assumed link flow distributions.
Some actual network analyses in addition to the simulation study are required to
support the result above.
55
REFERENCES
Abrahamsson, T., Estimation of OriginDestination Matrices Using Traffic Counts A
Literature Survey, International Institute for Applied Systems Analysis, May 1998
<URL: http://www.iiasa.ac.at/Publications/Documents/IR98021.pdf> [Jan. 3
rd
, 2002]
Brillouin, L., 1962. Science and Information Theory 2
nd
Edition. Academic Press Inc.
CORSIM User Manual: Version 1.01, 1996. Federal Highway Administration, US
Department of Transportation, Virginia, USA
<URL: http://plan2op.fhwa.dot.gov/pdfs/Pdf2/Edl06408.pdf > [Jan. 12
th
, 2002]
Dickey, J.W., Diewald , W.J., Hobeika, A.G., Hurst, C.J., Stephens, N.T., Stuart, R.C.,
Walker, R.D., 1983. Metropolitan Transportation Planning. Hemisphere, USA.
Drew, D.R., 1968. Traffic Flow Theory and Control. McGrawHill, USA.
Feller, W., 1966. An Introduction to Probability Theory and Its Applications Vol. 1
st
Edition. Wiley, USA.
Hellinga, B.R., Estimating Dynamic Origin Destination Demand from Link and Probe
Counts, Queens University, August 1994.
<URL: http://www.civil.uwaterloo.ca/bhelling/Publications/PhD%20Thesis.pdf > [Dec. 1
st
,
2001]
Hornbeck, R.W., 1975. Numerical Methods. PrenticeHall, Englewood Cliffs, New Jersey,
USA
Homburger, W.S., Transportation and Traffic Engineering Handbook  2
nd
Edition,
PrenticeHall, Englewood Cliffs, New Jersey, USA.
Irwin, N.A., Cube, H.G.V., 1962. Capacity Restraint in Multitravel Mode assignment
programs. Highway Research Board, Bulletin 347, pp.258289.
Law, A.M., Kelton, W.D., 1982. Simulation Modeling and Analysis. McGrawHill, USA
Mannering, F.L., Kilareski, W.P., 1998. Principle of Highway Engineering and Traffic
Analysis. 2
nd
Edition, John Wiley & Sons, USA.
Mathetharan, V., 1997, Determination of Highway Capacity on Uninterrupted Flow
Highway, AIT Thesis No. TE964, Asian Institute of Technology, Bangkok, Thailand.
McShane, W.R., Roess, R.P., 1990. Traffic Engineering. Prentice Hall, Englewood Cliff,
New Jersey, USA.
More, J.E., Kim, G., Cho, S., Hu, H., and Xu, R., Evaluating System ATMIS
Technologies via Rapid Estimation of Network Flows: Final Report, University of
Southern of California, August 1997.
<URL: http://www.path.berkeley.edu/Quarterly/General/mous.phtml> [Jan. 23
rd
, 2002]
56
Murthy, N., Application of Traffic Simulation Models in Transportation Planning, Design
and Operations, Korve Engineering Transportation
< URL: http://www.itcworld.com/library/Application%20of%20Traffic%20Simulation%
20Models.pdf> [Feb. 14
th
, 2002]
Nakatsuji, T., 2001. Traffic Simulation Techniques. Asian Center for Transportation
Studies, Thailand.
Paramahamsan, H. Fundamental Properties of Synthetic OD Generation Formulations
and Solutions, Virginia Tech., January 1999.
<URL: http://scholar.lib.vt.edu/theses/available/etd020599151231/> [Dec. 25
th
, 2001]
Pignataro, L. J., 1973. Traffic Engineering, Englewood Cliffs; PrenticeHall, USA.
Sheffi, Y., 1985. Urban Transportation Network: Equilibrium Analysis with Mathematical
Programming Methods. PrenticeHall, Englewood Cliff, USA.
Surasak, T., 1998, Development of A Delay Formula and An Optimal Cycle Time Formula
for Isolated Pre Signalized Intersections in Bangkok, Thailand, AIT Thesis No. TE975,
Asian Institute of Technology, Bangkok, Thailand.
Zuylen, H.J.V., and Willumsen, L.G., 1979. The Most Likely Trip Matrix Estimated from
Traffic Counts. Transportation Research B, 14B: 281 293.
Vythoulkas, P.C., 1990. A Dynamic Stochastic Assignment Model for the Analysis of
General Networks. Transportation Research  B, 24 B: 453469.
Wardrop, J.G., 1952. Some Theoretical Aspects of Road Traffic Research. Proc. Ins.
Civil Engineering, Part II, 1, pp.325378.
Yue, P., and Yu, L., Travel Demand Forecasting Models: A Comparison of EMME/2 and
QRS II Using a RealWorld Network, Texas Southern University, October 2000.
<URL: http://swutc.tamu.edu/Reports/1679011.pdf> [Dec. 12
th
, 2001]
57
Appendix
Simulation Outputs
58
EXISTING CONDITION

NETWORKWIDE AVERAGE STATISTICS

(Case 1)
TOTAL VEHICLE MILE = 42821.36 VEHICLEHOURS OF: MOVE TIME = 1427.38 , DELAY TIME = 3695.11 , TOTAL TIME = 5122.49
AVERAGE SPEED ( MPH)= 8.36 MOVE/TOTAL = .28 MINUTES/MILE OF: DELAY TIME = 5.18 , TOTAL TIME = 7.18
(Case 2)
TOTAL VEHICLE MILE = 44750.08 VEHICLEHOURS OF: MOVE TIME = 1491.67 , DELAY TIME = 3861.76 , TOTAL TIME = 5353.43
AVERAGE SPEED ( MPH)= 8.36 MOVE/TOTAL = .28 MINUTES/MILE OF: DELAY TIME = 5.18 , TOTAL TIME = 7.18
(Case 3)
TOTAL VEHICLE MILE = 45058.24 VEHICLEHOURS OF: MOVE TIME = 1501.94 , DELAY TIME = 3761.19 , TOTAL TIME = 5263.13
AVERAGE SPEED ( MPH)= 8.56 MOVE/TOTAL = .29 MINUTES/MILE OF: DELAY TIME = 5.01 , TOTAL TIME = 7.01
(Case 4)
TOTAL VEHICLE MILE = 44629.41 VEHICLEHOURS OF: MOVE TIME = 1487.65 , DELAY TIME = 3816.17 , TOTAL TIME = 5303.82
AVERAGE SPEED ( MPH)= 8.41 MOVE/TOTAL = .28 MINUTES/MILE OF: DELAY TIME = 5.13 , TOTAL TIME = 7.13
(Case 5)
TOTAL VEHICLE MILE = 44571.32 VEHICLEHOURS OF: MOVE TIME = 1485.71 , DELAY TIME = 3724.97 , TOTAL TIME = 5210.68
AVERAGE SPEED ( MPH)= 8.55 MOVE/TOTAL = .29 MINUTES/MILE OF: DELAY TIME = 5.01 , TOTAL TIME = 7.01
59
(Case 6)
TOTAL VEHICLE MILE = 44221.89 VEHICLEHOURS OF: MOVE TIME = 1474.06 , DELAY TIME = 3585.32 , TOTAL TIME = 5059.38
AVERAGE SPEED ( MPH)= 8.74 MOVE/TOTAL = .29 MINUTES/MILE OF: DELAY TIME = 4.86 , TOTAL TIME = 6.86
(Case 7)
TOTAL VEHICLE MILE = 46407.29 VEHICLEHOURS OF: MOVE TIME = 1546.91 , DELAY TIME = 4073.86 , TOTAL TIME = 5620.77
AVERAGE SPEED ( MPH)= 8.26 MOVE/TOTAL = .28 MINUTES/MILE OF: DELAY TIME = 5.27 , TOTAL TIME = 7.27
(Case 8)
TOTAL VEHICLE MILE = 44306.18 VEHICLEHOURS OF: MOVE TIME = 1476.87 , DELAY TIME = 3673.62 , TOTAL TIME = 5150.50
AVERAGE SPEED ( MPH)= 8.60 MOVE/TOTAL = .29 MINUTES/MILE OF: DELAY TIME = 4.97 , TOTAL TIME = 6.97
(Case 9)
TOTAL VEHICLE MILE = 49401.00 VEHICLEHOURS OF: MOVE TIME = 1646.70 , DELAY TIME = 3954.70 , TOTAL TIME = 5601.40
AVERAGE SPEED ( MPH)= 8.82 MOVE/TOTAL = .29 MINUTES/MILE OF: DELAY TIME = 4.80 , TOTAL TIME = 6.80
(Case 10)
TOTAL VEHICLE MILE = 44190.34 VEHICLEHOURS OF: MOVE TIME = 1473.01 , DELAY TIME = 3728.23 , TOTAL TIME = 5201.24
AVERAGE SPEED ( MPH)= 8.50 MOVE/TOTAL = .28 MINUTES/MILE OF: DELAY TIME = 5.06 , TOTAL TIME = 7.06
60
FIRST APPROACH

NETWORKWIDE AVERAGE STATISTICS

(Case 1)
TOTAL VEHICLE MILE = 41921.43 VEHICLEHOURS OF: MOVE TIME = 1397.38 , DELAY TIME = 3406.45 , TOTAL TIME = 4803.83
AVERAGE SPEED ( MPH)= 8.73 MOVE/TOTAL = .29 MINUTES/MILE OF: DELAY TIME = 4.88 , TOTAL TIME = 6.88
(Case 2)
TOTAL VEHICLE MILE = 39502.44 VEHICLEHOURS OF: MOVE TIME = 1316.75 , DELAY TIME = 3480.73 , TOTAL TIME = 4797.47
AVERAGE SPEED ( MPH)= 8.23 MOVE/TOTAL = .27 MINUTES/MILE OF: DELAY TIME = 5.29 , TOTAL TIME = 7.29
(Case 3)
TOTAL VEHICLE MILE = 38345.16 VEHICLEHOURS OF: MOVE TIME = 1278.17 , DELAY TIME = 3323.76 , TOTAL TIME = 4601.93
AVERAGE SPEED ( MPH)= 8.33 MOVE/TOTAL = .28 MINUTES/MILE OF: DELAY TIME = 5.20 , TOTAL TIME = 7.20
(Case 4)
TOTAL VEHICLE MILE = 38720.95 VEHICLEHOURS OF: MOVE TIME = 1290.70 , DELAY TIME = 3503.83 , TOTAL TIME = 4794.53
AVERAGE SPEED ( MPH)= 8.08 MOVE/TOTAL = .27 MINUTES/MILE OF: DELAY TIME = 5.43 , TOTAL TIME = 7.43
(Case 5)
TOTAL VEHICLE MILE = 39122.96 VEHICLEHOURS OF: MOVE TIME = 1304.10 , DELAY TIME = 3263.40 , TOTAL TIME = 4567.49
AVERAGE SPEED ( MPH)= 8.57 MOVE/TOTAL = .29 MINUTES/MILE OF: DELAY TIME = 5.00 , TOTAL TIME = 7.00
61
(Case 6)
TOTAL VEHICLE MILE = 39462.84 VEHICLEHOURS OF: MOVE TIME = 1315.43 , DELAY TIME = 3382.50 , TOTAL TIME = 4697.93
AVERAGE SPEED ( MPH)= 8.40 MOVE/TOTAL = .28 MINUTES/MILE OF: DELAY TIME = 5.14 , TOTAL TIME = 7.14
(Case 7)
TOTAL VEHICLE MILE = 41564.45 VEHICLEHOURS OF: MOVE TIME = 1385.48 , DELAY TIME = 3796.23 , TOTAL TIME = 5181.71
AVERAGE SPEED ( MPH)= 8.02 MOVE/TOTAL = .27 MINUTES/MILE OF: DELAY TIME = 5.48 , TOTAL TIME = 7.48
(Case 8)
TOTAL VEHICLE MILE = 38735.65 VEHICLEHOURS OF: MOVE TIME = 1291.19 , DELAY TIME = 3297.15 , TOTAL TIME = 4588.34
AVERAGE SPEED ( MPH)= 8.44 MOVE/TOTAL = .28 MINUTES/MILE OF: DELAY TIME = 5.11 , TOTAL TIME = 7.11
(Case 9)
TOTAL VEHICLE MILE = 38935.37 VEHICLEHOURS OF: MOVE TIME = 1297.85 , DELAY TIME = 3214.18 , TOTAL TIME = 4512.02
AVERAGE SPEED ( MPH)= 8.63 MOVE/TOTAL = .29 MINUTES/MILE OF: DELAY TIME = 4.95 , TOTAL TIME = 6.95
(Case 10)
TOTAL VEHICLE MILE = 38703.62 VEHICLEHOURS OF: MOVE TIME = 1290.12 , DELAY TIME = 3317.39 , TOTAL TIME = 4607.51
AVERAGE SPEED ( MPH)= 8.40 MOVE/TOTAL = .28 MINUTES/MILE OF: DELAY TIME = 5.14 , TOTAL TIME = 7.14
62
SECOND APPROACH

NETWORKWIDE AVERAGE STATISTICS

(Case 1)
TOTAL VEHICLE MILE = 42847.82 VEHICLEHOURS OF: MOVE TIME = 1448.97 , DELAY TIME = 3457.22 , TOTAL TIME = 4906.18
AVERAGE SPEED ( MPH)= 8.86 MOVE/TOTAL = .30 MINUTES/MILE OF: DELAY TIME = 4.77 , TOTAL TIME = 6.77
(Case 2)
TOTAL VEHICLE MILE = 44010.46 VEHICLEHOURS OF: MOVE TIME = 1467.02 , DELAY TIME = 3494.83 , TOTAL TIME = 4961.85
AVERAGE SPEED ( MPH)= 8.87 MOVE/TOTAL = .30 MINUTES/MILE OF: DELAY TIME = 4.76 , TOTAL TIME = 6.76
(Case 3)
TOTAL VEHICLE MILE = 44620.37 VEHICLEHOURS OF: MOVE TIME = 1487.35 , DELAY TIME = 3226.29 , TOTAL TIME = 4713.64
AVERAGE SPEED ( MPH)= 9.47 MOVE/TOTAL = .32 MINUTES/MILE OF: DELAY TIME = 4.34 , TOTAL TIME = 6.34
(Case 4)
TOTAL VEHICLE MILE = 43831.17 VEHICLEHOURS OF: MOVE TIME = 1461.04 , DELAY TIME = 3400.72 , TOTAL TIME = 4861.76
AVERAGE SPEED ( MPH)= 9.02 MOVE/TOTAL = .30 MINUTES/MILE OF: DELAY TIME = 4.66 , TOTAL TIME = 6.66
(Case 5)
TOTAL VEHICLE MILE = 41521.48 VEHICLEHOURS OF: MOVE TIME = 1384.05 , DELAY TIME = 3198.56 , TOTAL TIME = 4582.61
AVERAGE SPEED ( MPH)= 9.06 MOVE/TOTAL = .30 MINUTES/MILE OF: DELAY TIME = 4.62 , TOTAL TIME = 6.62
63
(Case 6)
TOTAL VEHICLE MILE = 43101.40 VEHICLEHOURS OF: MOVE TIME = 1436.71 , DELAY TIME = 3120.04 , TOTAL TIME = 4556.76
AVERAGE SPEED ( MPH)= 9.46 MOVE/TOTAL = .32 MINUTES/MILE OF: DELAY TIME = 4.34 , TOTAL TIME = 6.34
(Case 7)
TOTAL VEHICLE MILE = 44831.46 VEHICLEHOURS OF: MOVE TIME = 1494.38 , DELAY TIME = 3429.30 , TOTAL TIME = 4923.69
AVERAGE SPEED ( MPH)= 9.11 MOVE/TOTAL = .30 MINUTES/MILE OF: DELAY TIME = 4.59 , TOTAL TIME = 6.59
(Case 8)
TOTAL VEHICLE MILE = 44160.82 VEHICLEHOURS OF: MOVE TIME = 1472.03 , DELAY TIME = 3454.96 , TOTAL TIME = 4926.99
AVERAGE SPEED ( MPH)= 8.96 MOVE/TOTAL = .30 MINUTES/MILE OF: DELAY TIME = 4.69 , TOTAL TIME = 6.69
(Case 9)
TOTAL VEHICLE MILE = 40728.05 VEHICLEHOURS OF: MOVE TIME = 1357.60 , DELAY TIME = 3181.47 , TOTAL TIME = 4539.07
AVERAGE SPEED ( MPH)= 8.97 MOVE/TOTAL = .30 MINUTES/MILE OF: DELAY TIME = 4.69 , TOTAL TIME = 6.69
(Case 10)
TOTAL VEHICLE MILE = 43994.90 VEHICLEHOURS OF: MOVE TIME = 1507.93 , DELAY TIME = 3456.71 , TOTAL TIME = 4964.64
AVERAGE SPEED ( MPH)= 9.11 MOVE/TOTAL = .30 MINUTES/MILE OF: DELAY TIME = 4.58 , TOTAL TIME = 6.58
64
THIRD APPROACH

NETWORKWIDE AVERAGE STATISTICS

(Case 1)
TOTAL VEHICLE MILE = 41364.15 VEHICLEHOURS OF: MOVE TIME = 1378.80 , DELAY TIME = 3180.98 , TOTAL TIME = 4559.79
AVERAGE SPEED ( MPH)= 9.07 MOVE/TOTAL = .30 MINUTES/MILE OF: DELAY TIME = 4.61 , TOTAL TIME = 6.61
(Case 2)
TOTAL VEHICLE MILE = 43327.54 VEHICLEHOURS OF: MOVE TIME = 1444.25 , DELAY TIME = 3047.35 , TOTAL TIME = 4491.60
AVERAGE SPEED ( MPH)= 9.65 MOVE/TOTAL = .32 MINUTES/MILE OF: DELAY TIME = 4.22 , TOTAL TIME = 6.22
(Case 3)
TOTAL VEHICLE MILE = 42178.09 VEHICLEHOURS OF: MOVE TIME = 1405.94 , DELAY TIME = 3087.11 , TOTAL TIME = 4493.05
AVERAGE SPEED ( MPH)= 9.39 MOVE/TOTAL = .31 MINUTES/MILE OF: DELAY TIME = 4.39 , TOTAL TIME = 6.39
(Case 4)
TOTAL VEHICLE MILE = 41535.96 VEHICLEHOURS OF: MOVE TIME = 1384.53 , DELAY TIME = 2951.60 , TOTAL TIME = 4336.13
AVERAGE SPEED ( MPH)= 9.58 MOVE/TOTAL = .32 MINUTES/MILE OF: DELAY TIME = 4.26 , TOTAL TIME = 6.26
(Case 5)
TOTAL VEHICLE MILE = 42384.21 VEHICLEHOURS OF: MOVE TIME = 1412.81 , DELAY TIME = 3031.99 , TOTAL TIME = 4444.80
AVERAGE SPEED ( MPH)= 9.54 MOVE/TOTAL = .32 MINUTES/MILE OF: DELAY TIME = 4.29 , TOTAL TIME = 6.29
65
(Case 6)
TOTAL VEHICLE MILE = 42626.46 VEHICLEHOURS OF: MOVE TIME = 1420.88 , DELAY TIME = 3025.25 , TOTAL TIME = 4446.13
AVERAGE SPEED ( MPH)= 9.59 MOVE/TOTAL = .32 MINUTES/MILE OF: DELAY TIME = 4.26 , TOTAL TIME = 6.26
(Case 7)
TOTAL VEHICLE MILE = 43461.71 VEHICLEHOURS OF: MOVE TIME = 1448.72 , DELAY TIME = 3239.28 , TOTAL TIME = 4688.00
AVERAGE SPEED ( MPH)= 9.27 MOVE/TOTAL = .31 MINUTES/MILE OF: DELAY TIME = 4.47 , TOTAL TIME = 6.47
(Case 8)
TOTAL VEHICLE MILE = 41672.89 VEHICLEHOURS OF: MOVE TIME = 1389.10 , DELAY TIME = 3028.58 , TOTAL TIME = 4417.67
AVERAGE SPEED ( MPH)= 9.43 MOVE/TOTAL = .31 MINUTES/MILE OF: DELAY TIME = 4.36 , TOTAL TIME = 6.36
(Case 9)
TOTAL VEHICLE MILE = 41719.42 VEHICLEHOURS OF: MOVE TIME = 1390.65 , DELAY TIME = 2941.34 , TOTAL TIME = 4331.99
AVERAGE SPEED ( MPH)= 9.63 MOVE/TOTAL = .32 MINUTES/MILE OF: DELAY TIME = 4.23 , TOTAL TIME = 6.23
(Case 10)
TOTAL VEHICLE MILE = 39487.91 VEHICLEHOURS OF: MOVE TIME = 1316.26 , DELAY TIME = 2769.49 , TOTAL TIME = 4085.75
AVERAGE SPEED ( MPH)= 9.66 MOVE/TOTAL = .32 MINUTES/MILE OF: DELAY TIME = 4.21 , TOTAL TIME = 6.21