Lb/noH/£Z/ob 
A REAL TIME TRAFFIC SIGNAL CONTROL 
SYSTEM 
A dissertation submitted to the 
Department of Electrical Engineering, University of Moratuwa 
in partial fulfillment of the requirements for the 
degree of Master of Science 
LIBRARY 
UNIVERSITY OF MORATUWA, SRI LANKA 
M O R A T U W A . 
by 
JAYAKODY ARACHCHIGE NIMANTHI NISHANI 
KUMUDU JAYAKODY 
Supervised by: Dr. Lanka Udawatta and Dr. Sisil Kumarawadu, 
Department of Electrical Engineering 
University of Moratuwa, Sri Lanka 
University of Moratuwa 
91207 
g\zo 
January 2008 
9 1 2 0 7 i 
DECLARATION 
The work submitted in this dissertation is the result of my own 
investigation, except where otherwise stated. 
It has not already been accepted for any degree, and is also not being 
concurrently submitted for any other degree. 
J.A.N.N.K Jayakody 
25.01.2008 
We endorse the declaration by the candidate. 
Dr. Lanka Udawatta 
skJ "A 
Dr. Sisil Kumarawadu 
11 
CONTENTS 
Declaration 
Abstract 
Dedication 
Acknowledgement 
List of Figures 
List of Tables 
Chapter 1 Introduction 
1.1 Background 
1.2 Motivation 
1.3 Literature Review 
1.4 Contributions of the Research 
1.5 Organization of the Report 
Chapter 2 Available Adaptive Traffic Control Systems 
2.1 Introduction 
2.2 Major Methodologies 
2.3 Split Cycle Offset Optimization Technique 
2.4 Sydney Coordinated Adaptive Traffic System 
2.5 Real time Hierarchical Optimized Distributed & Effective 
System 
2.6 OP AC - Optimized Poicies for Adaptive Control 
Chapter 3 Decentralized Intelligent control Model i s 
3.1 Introduction ^ 
3.2 Detection in Adaptive Control 1 g 
3.2.1 Function of Detection j g 
3.2.2 Detection Methods j9 
iii 
3.2.3 Inductive Loop Detectors 20 
3.3 Influence Detection 21 
3.3.1 Structure of the Intelligent Control 21 
3.3.2 Formulating the Influence Function for Traffic Signal Control 22 
Chapter 4 Fuzzy Control Strategy Development 25 
4.1 Introduction 25 
4.2 Fuzzy Inference System 25 
4.2.1 Fuzzy Basics 25 
4.2.2 Fuzzification 26 
4.2.3 Rule Base 3 0 
4.2.4 Determination of the Single Output 33 
4.3 Control Strategy 34 
4.3.1 System Architecture 34 
4.3.2 Decentralized Approach 35 
4.3.3 Prediction Method 36 
4.3.4 Control Mechanism 37 
Chapter 5 Simulation and Results 41 
5.1 Introduction 4] 
5.2 Simulation Model Development 41 
5.2.1 Simulink Basics 41 
5.2.2 Traffic Simulation Model of an Intersection 42 
5.2.3 Simulation Results 44 
5.3 Comparison with existing Fixed time Traffic Control System 46 
5.3.1 Delay Calculation 46 
5.3.2 Delay Calculation for the Approach A 47 
IV 
5.3.3 Delay Calculation for the Approach B 
5.3.4 Delay Calculation for the Approach C 
5.3.5 Delay Calculation for the Approach D 
49 
51 
53 
Chapter 6 
References 
Conclusions and Future work 
6.1 Conclusions 
6.2 Future Work 
56 
56 
56 
58 
v 
Abstract 
Traffic congestion problem in Colombo city is getting worse since traditional traffic 
control system could not fulfill the need. Since the existing system is a fixed time 
fixed cycle control system, it cannot fit with dynamic traffic environment. 
In this research, a decentralized control strategy to control a traffic network grid is 
presented. Single controller is to control traffic signals of all approaches at one 
intersection and each approach green time is given by its separate Fuzzy Inference 
System. Vehicle arrival data are to be collected by lane detectors. Inductive Loop 
Detectors are proposed for this purpose. Herein, a methodology is developed to 
decide green time of each approach based on the arrival data by the Fuzzy Inference 
System and the Cycle time. Influence to the particular intersection is identified and is 
factorized as an input to the Fuzzy Inference System. Later, the green time is decided 
by the FIS. Results for this mechanism are shown for one intersection on a simulated 
environment modeled by Matlab. 
Calculations have been done based on the real data obtained for fifteen occasions. 
Results for three sets of data from both existing fixed time system and the intelligent 
model have been compared based on the calculations done for the total vehicle delay 
time, expected at the passing the particular intersection. It shows 51.6% of minimized 
total vehicle seconds delay by the intelligent traffic control model over the fixed time 
control system. 
vi 
To my parents 
vii 
Acknowledgement 
This is to pay my warm gratitude to the Department of Electrical Engineering, 
University of Moratuwa, SriLanka for giving me the opportunity to read for the M.Sc 
in Industrial Automation and to do this research during the period of October 2005 to 
January 2008. 
I would like to extend my warm thanks with respect to my research supervisors Dr. 
Lanka Udawatta and Dr. Sisil Kumarawadu, who have given endless support and 
guidance while I was doing the research and following the program. Their breadth of 
knowledge and the advices given me at the presentations made me highly impressed 
to do the research in depth and to develop myself by enriching academic, research 
oriented and professional maturity. Their countless advices and the guidance were 
invaluable. 
I would also like to thank Professor Ranjith Perera, the Head of the Department. He 
has given me support and encouragement, and his advice and feedback about my 
research at presentations have greatly enhanced and strengthened the study. I thank 
him for all the time and energy he has paid for my work. 
I also wish to extend thanks to my friends and colleagues who were invaluable in 
completing this study, and to the management of Electro-Serv (pvt) ltd who has given 
me the support by releasing me from the duty to follow this program. 
I am indebted to my parents and to my brother Suranga Jayakody for the guidance 
they have given to me. I recall their constant support, encouragement, love and 
guidance given me every time. 
Finally, I would like to share my research experience with all of you. 
Nimanthi Jayakody 
University of Moratuwa, Sri Lanka 
January 2008 
viii 
List of Figures 
Figure Page 
3.1 Block diagram of the Inductive Loop Detector 20 
3.2 Part of the traffic network 22 
3.3 Intersection A 23 
4.1 Trapezoidal membership function for the input Influence 27 
4.2 Triangular membership function for the output Split 28 
4.3 Variation of Split Vs. Influence 29 
4.4 Graphical representation of rule evaluation 31 
4.5 Block diagram of the control system 35 
4.6 Master Slave control model 36 
4.7 Flow chart for proposed adaptive control mechanism 39 
5.1 Traffic simulation model developed by Matlab 43 
5.2 Arrivals at an intersection over the cycle time 47 
IV 
List of Tables 
Table Page 
4.1 Membership Values of the Input Influence 27 
4.2 Membership Values of the Output Split 28 
5.1 Detector Counts at Approach Counters 44 
5.2 Influences calculated for the detector data 44 
5.3 Splits determined by the Fuzzy Inference Systems 45 
5.4 Delay Calculation for the Intelligent Model at Split A 48 
5.5 Delay Calculation for the Fixed time Control Model at 
Split A 49 
5.6 Delay Calculation for the Intelligent Model at Split B 50 
5.7 Delay Calculation for the Fixed time Control Model at 
Split B 50 
5.8 Delay Calculation for the Intelligent Model at Split C 52 
5.9 Delay Calculation for the Fixed time Control Model at 
Split C 52 
5.10 Delay Calculation for the Intelligent Model at Split D 54 
5.11 Delay Calculation for the Fixed time Control Model at 
Split D 54 
5.12 Improvement obtained by the Intelligent Control Model 55 
x 
Chapter 1: Introduction 
1.1 Background 
The problem of controlling traffic signals, which is the problem of deciding when to 
allow vehicles to cross an intersection, has been dealt in many ways. Dramatic 
progress has been achieved from the traffic guide, who would direct drivers at 
intersections with a flag to the implementation of fully actuated signal control systems 
to the present day systems of interconnected signals in corridors. The availability of 
detectors of many kinds; e.g. loop detection, video detection, and infrared detection, 
that are capable of transmitting information on the real traffic situation, and the 
development of microprocessors which can process this data, motivates the 
development of real-time traffic control systems. 
Urban vehicular traffic is an expression of human activity and is highly variable in 
time and space. Traffic load is highly dependent on parameters such as time, day, 
season, weather and unpredictable situations. If these parameters are not taken in to 
account, the traffic control system will create bottlenecks and delays. Therefore, the 
control of such traffic requires a high degree of adaptability to produce suitable 
responses to this variability. Existing system (fixed time control system) however, 
impose certain rigidities that very much restrict the opportunity to get adapted with 
the real time traffic load. 
A traffic intersection is defined by a location in which the sharing of right of way by 
two or more traffic streams is required. In order to accomplish this sharing, 
intersection control is used. The traffic signal operates by allotting green time to the 
intersection approaches by a predefined way. The manner in which green time is 
allotted to these approaches is categorized as follows [3], [9], 
(a) Pretimed or Fixed: The traffic signal provides a fixed amount of green time to each 
approach during a cycle. This green duration is fixed for each interval for some period 
of time. 
(b) Actuated: The traffic signal provides a minimum length of green time to each 
approach during a cycle. This length may be incremented based on vehicle arrivals to 
the approach displaying green as observed by a detection device. The length of every 
green interval is also constrained by a maximum green time specification. 
1 
(c) Adaptive: The traffic signal provides green time to each intersection approach 
based on anticipated arrivals for a cycle. Generally, as arrivals change from cycle to 
cycle, the length of green time provided to each approach also changes. 
In intelligent traffic control systems reviews, the main terminologies found are split, 
cycle and offset [29], The most existing adaptive methods are to optimize these three 
factors. The cycle time means the total time allowed for all approaches to control the 
traffic flow at the particular intersection. The split is meant as the green phase or the 
fractions of the cycle time given to each approach or the green time at the approach. 
Offset is the delay between two neighboring intersections to initiate each cycle. 
Fixed time signal means that controlling the traffic light by predefined timetables [7]. 
The traffic signal provides a fixed amount of green time to each approach during a 
cycle. This green duration is fixed for each interval for some period of time. In traffic 
actuated signal, cycles, signal phases and signal intervals of the traffic light are 
defined in controllers. Conventional traffic-control strategies have limitations in 
handling unanticipated traffic demands. In coordinated systems, whether they are pre-
timed or coordinated actuated, engineers use historical traffic data to develop a timing 
plan that would optimize operation of the intersection [9], 
An adaptive traffic-signal control on the other hand collects traffic data in real-time, 
and therefore is expected to mitigate the problem of using outdated data and to 
improve overall system performance.- While an improvement in average delay 
throughout the system and lower travel time through the corridor are expected to be 
the main outcomes of an adaptive system, this improvement is achieved at the 
expense of extra delays for side street users. In network based control systems, the 
sensors will send monitored traffic data to master controller and according to those 
values, the master controller calculate and periodically distribute appropriate traffic 
signal control strategy to the controllers at intersections [19]. In decentralized control 
system, individual controllers implement its own control strategy. 
2 
1.2 Motivation 
Today in Srilanka, the traffic congestion in urban area especially in Colombo is an 
important issue to be solved. One of the reasons of this congestion is imperfect 
operation of traffic signal control. 
In Colombo, it is used fixed cycle, fixed time based control system. Being not get 
adapted to the actual traffic load, it results for heavy traffic congestion on roads at 
peak hours, most of the times in morning and in evening when people get rush for 
work and school. In some junctions, traffic control by a police officer still exists. 
Therefore, a proper coordination among junctions cannot be expected and it results for 
increasing the delay at intersections. 
The objective of the implementation of an Intelligent Control System is to minimize 
the average total delay for all vehicles entering the intersection. It provides many 
advantages over existing fixed-time plans, including minimizing traffic delays by 
providing effective signal operation, improving travel time, reducing fuel 
consumption and improving air quality and reducing maintenance and operations cost. 
The goal of this research is to study and apply traffic control system that operates in 
real time, adjusting signal timing to accommodate changing traffic patterns. It is not 
based on a fixed cycle length. By an intelligent traffic control system, the green time 
is provided to each intersection approach based on anticipated arrivals for a cycle. 
Generally, as arrivals change from cycle to cycle, the length of green time provided to 
each approach also changes. In this study, it is proposed a simulation model for a 
fuzzy controlled decentralized traffic control system to improve existing system and 
to reduce total delay time. 
1.3 Literature review 
Over the course of performing a literature review, it was found that over the past 20 to 
30 years, research in the area of adaptive signal control has gradually been increasing 
and practical applications are beginning to be seen around the world. Multi - agent 
system to urban traffic signal control by Szu-Yin Lin., et al[l] has shown an agent 
model for a decentralized traffic signal control. Development of an autonomous 
adaptive traffic control system by K. Tavladaskis[2] shows a network approach. 
3 
Researchers have offered the definitions for the parameters and measures of 
effectiveness in various ways. In this research, the definitions and the parameters 
given by Szu-Yin Lin, et al have been used. 
The world popular traffic control methods are Split Cycle Offset Optimization 
Technique SCOOT, Sydney Coordinated Adaptive Traffic Control System SCATS, 
Real time Hierarchical Optimized Distributed and Effective System RHODES and the 
Optimized Policy for Adaptive Control OP AC. Here SCOOT and SCATS use a cycle 
based approach on a network while RHODES and OPAC work on the concept of 
rolling horizon approach. The TRYS approach by Josefa Hernandez[4], and RHODES 
by Pitu Mirchandani., et al[5] have shown more hierarchical manner to the adaptive 
control system. . The web page http://www.scoot-utc.com/fl4] gives a complete 
picture of SCOOTS and the places it has been installed. RHODES by Pitu 
Mirchandani, et al[5] have discussed the control system RHODES. The architecture, 
which decomposes the traffic control problem to several sub problems, predicts traffic 
flows at appropriate resolution levels and the optimization modules to solve the 
hierarchical sub problems while utilizing a data structure for fast solution of the sub 
problems and the analysis are presented. The web page 
http://ops.fhwa.dot.gov/publications/adaptivecontrol/ [25] discusses about OPAC and 
http://www.traffic-tech.com/pdf/scatsbrochure.pdf [30] discusses SCATS control 
system in detail. 
A micro simulation model, SIGSIM for-simulating movement of individual vehicles 
in a signal controlled road network is presented in Adaptive traffic control using 
evolutionary algorithms by Jenan Sha Aban, et al [6], Results obtained for mean rates 
of delay over simulation run is also given. 
Josefa Hernandez, et al have discussed the TRYS model, a good way of presenting the 
traffic control model in Real time traffic Management through knowledge based 
models: TRYS approach [4], A set of data bases, each one describing the physical 
structure of a problematic area, a collection of frames representing demand models of 
different problem areas and the knowledge base to support control decisions 
organized as a collection of control frames are included in the TRYS approach. 
Michael and Randy Machemehl[3] have shown more experiment oriented delay 
optimized approach and the delays have been calculated. 
4 
1.4 Contributions of the research 
This project is investigating a method of adaptive traffic signal control over the real 
time approach, which will meet the changing traffic needs. It has a decentralized 
control strategy. It was designed to suit for an intersection in Colombo city. The 
controller is to be installed at the junction and control strategy is to control the traffic 
flow at that junction only. Since each junction has its own controller, the need for a 
central controller is not needed. 
Single input single output Fuzzy Inference System is to decide the green time of that 
particular approach and all four splits are totalized to calculate the cycle time. The 
opening of each approach at the given cycle is taken place starting from the north 
approach and is ended at West. The outputs of the controller are energized following 
that mentioned sequence. A simulation model was developed to meet the above 
design. Simulation results for a set of data and the comparison with the existing fixed 
time control system are also provided. 
1.5 Organization of the report 
In this chapter, a focus of the concept development is introduced. In chapter 2, four 
most popular adaptive traffic control systems are described from reviewing the 
previous work carried out in this topic. In chapter 2, the way of detection, the way of 
getting influence and the effect of detection values to the influence are discussed. 
Chapter 3 presents the Fuzzy Control Strategy development in the research. The way 
of control mechanism developed and the way of controller perform are discussed in 
this chapter. Chapter 4 provides the simulation results for a set of cases and a 
comparison of the intelligent control model with the existing fixed time control 
system is also provided. 
5 
Chapter 2: Available Adaptive traffic control systems 
2.1 Introduction: 
Over the course of performing a literature review, it was found that there are four 
methodologies that stand out from other attempts at adaptive signal control. They are 
significant due to their relative acceptance in the field, as well as the relative extent of 
their real- world implementation. In this chapter, the most popular adaptive traffic 
control systems are discussed. 
2.2 Major methodolies: 
There are many adaptive systems available but many have been developed specifically 
for a particular city and whilst they may be available for use by others, they do not 
have the support and ongoing development opportunities afforded by more 
commercially available systems. The most widely used among these, and available in 
a relatively off-the-shelf deployment solution are the British SCOOT (Split Cycle 
Offset Optimization Technique) [14] and the Australian SCATS (Sydney Coordinated 
Adaptive Traffic System) systems [28], In addition to them, RHODES (Real time 
Hierarchical Optimized Distributed and Effective system) [5] and the OPAC 
(Optimized Policies for Adaptive Control) [30] are used. 
2.3 SCOOT (Split Cycle Offset Optimization Technique): 
SCOOT was developed in the early 1980's by the transport research laboratory in the 
United Kingdom. For the detection method, inductive loops are most common, though 
other types of detectors can be used. For best results, detectors are required on each 
link. Installing inductive loops, and maintaining them subsequently, is a significant 
element in the cost of SCOOT, although less than would be required if all the 
junctions were operated by isolated Vehicle Actuation. Overhead detectors have been 
used successfully in some situations [14]. 
SCOOT differs from SCATS in that it uses a second set of advance vehicle detectors 
typically 50-300 meters upstream of the stop line. The advance detectors provide a 
count of the vehicles approaching at each junction. This gives the system a higher 
resolution picture of traffic flows and a count of the number of vehicles in each queue, 
6 
several seconds before they touch the stop line. It also provides exceptional queue 
length detection information to the system, which is triggered when the traffic queue 
backs up to the upstream detector. Under the SCOOT system green waves can be 
dynamically delayed on a just in time basis based on the arrival of vehicles at the 
upstream detector, allowing extra time to the previous green phase where warranted in 
heavy traffic conditions. 
A SCOOT network is divided into "regions", each containing a number of "nodes" 
(signalled junctions and pedestrian crossings) that all run at the same cycle time to 
allow co-ordination [14, 25, 27], 
Nodes may be "double cycled" (i.e. operate at half of the regional cycle time) at 
pedestrian crossings or under saturated junctions. Region boundaries are located 
across links where co-ordination is least critical, e.g. long links. Data on the regions, 
nodes, stages, links and detectors will need to be stored in the SCOOT database. 
When all the equipment has been installed and the network data input into the 
database, the system will need to be validated. Validation of SCOOT is the process of 
calibrating the SCOOT traffic model so that it reflects as accurately as possible the 
actual events on the street network. This is critical, to ensure effective performance of 
the system. Those parts of the system that have been validated can be operated under 
SCOOT control whilst further nodes are being validated. Once the system has been 
validated, the traffic management parameters can be set to manage traffic in line with 
the authority's strategy [14], 
When vehicles pass the detector, SCOOT receives the information and converts the 
data into its internal units and uses them to construct "Cyclic flow profiles" for each 
link. Vehicles are modeled down the link at cruise speed and join the back of the 
queue (if present). During the green, vehicles discharge from the stop line at the 
validated saturation flow rate. 
The data from the model is then used by SCOOT in three optimizers which are 
continuously adapting three key traffic control parameters - the amount of green for 
each approach (Split), the time between adjacent signals (Offset) and the time allowed 
for all approaches to a signaled intersection (Cycle time)[14]. 
7 
At every junction and for every phase, the split optimizer, a few seconds before the 
phase change is due to take place, will make a decision as to whether to make the 
change earlier, later or as due. The optimizer implements the decision, which only 
affects the phase change time by a few seconds that minimizes the maximum degree 
of saturation for the approaches to the intersection. 
During a predetermined phase in each cycle, and for every junction in the system, the 
offset optimizer makes a decision to alter, by a fixed amount, all the scheduled change 
times. The optimizer uses information stored in the cyclic flow profiles and by 
comparing the sum of performance indices on all the adjacent LINKs for the 
scheduled offset and the possible changed offsets. 
A SCOOT system is split into cycle time regions which have pre-determined 
minimum and maximum boundaries. The Cycle time Optimizer can vary the cycle 
time of each region in small intervals at attempt to ensure that the most heavily loaded 
node in the system is operating at a 90% saturation. If all stop bars are operating at 
less than 90% then the optimizer will make incremental reductions in cycle time. 
These three optimizers are used to continuously adapt these parameters for all 
intersections in the SCOOT controlled area, minimizing wasted green time at 
intersections and reducing stops and delays by synchronizing adjacent sets of signals. 
This means that signal timings evolve as the traffic situation changes without any of 
the harmful disruption caused by changing fixed time plans on more traditional urban 
traffic control systems [27], 
If one takes the example of a busy arterial road with side roads intersecting it, a 
SCOOT controlled traffic signal system will seek to maximize the traffic carrying 
capacity of the arterial road by slotting traffic flows from the side road, far side turn 
filters, and pedestrian crossing phases into predicted gaps in vehicle flows along the 
arterial route. The SCOOT system uses upstream vehicle detectors to predict vehicle 
arrivals at the stop line, in advance of real time. Upstream detectors on the side roads 
give the system information on queue lengths, which in turn allows it to maximize the 
green wave along the arterial route, by recovering un-needed green time allocated to 
non-priority phases. As AM and PM traffic peaks, SCOOT increases the cycle time 
8 
dramatically to assist traffic flow fluidity along main routes by reducing the incidence 
of start/stop events, which might otherwise bring traffic to a standstill. 
The SCOOT system is used in many other urban areas, including Toronto, San Diego, 
Anaheim, London and Bangkok. 
2.4 SCATS - Sydney Coordinated Adaptive Traffic System: 
SCATS is somewhat newer, having being created in the early 1990's by the Roads 
and Traffic Authority of New South Wales, Austrailia. SCATS is a proven adaptive 
system that is currently controlling more than 20,000 intersections on six continents 
[28], SCATS's ability to improve traffic signal operations and provide benefits to the 
traveling public has been proven in deployments across the United States [30], 
SCATS gathers data on traffic flows in real time at each intersection. This data is fed 
via the traffic control signal box to a central computer. The computer makes 
incremental adjustments to traffic light timings based on minute by minute changes in 
traffic flow at each intersection. SCATS performs a vehicle count at each stop line, 
and also measures the gap between vehicles as they pass through each junction. As the 
gap between vehicles increases, the lights are wasting green time and SCATS seeks to 
reallocate green time to where demand is- greatest. The SCATS system is also used in 
many other urban areas including Hong Kong, Sydney, Melborne and Oakland 
Country. 
Utilizing the Microsoft Windows Operating System, SCATS offers intuitive and 
efficient system operations and data entry [28], Using contemporary off-the-shelf 
servers and workstations, SCATS's scalable architecture is suitable for systems 
ranging from 10 intersections to over 2,000 intersections. A single server can control 
up to 250 intersections and additional servers can be added to provide control for more 
intersections. Its client-server architecture is suitable for conventional office networks, 
and its low processing requirements allows SCATS to be installed on typical office 
computers. SCATS provides its users contemporary, non-proprietary controller 
9 
options including support for Type 170-E, Model 2070, and Eagle M50 series 
controllers. 
SCATS also provide a flexible communications architecture, supporting: 
Point-to-point communications 
_ Point-to-multipoint communications 
_ IP-based communications 
Its robust communications protocol operates over fiberoptic networks, wireless 
networks, and twisted pair networks found in many municipalities. 
The proposed system has two control levels; strategic, and tactical. The Strategic 
Level establishes limits on the operation of control at each intersection. The Tactical 
Level Control resides at the controller, and provides for cycle-by-cycle adjustments 
within limits set by the Strategic Level [7, 24], 
Combined, both control levels will provide three main elements: 
• Cycle length calculations for each critical intersection; 
• Phase split calculation for each intersection; and 
• Determination of offsets between adjacent intersections. 
By design, the Cupertino city arterials network to be coordinated is subdivided into 
small groups, and each group is having a maximum of one critical intersection. A 
group may be any number of intersections, but is always be composed of at least one 
v. 
intersection. One key function of the Strategic Control is to decide when to join two 
or more groups into one coordinated systems. The desired cycle lengths for individual 
groups are used for this purpose. When two groups require similar cycle lengths, they 
could be joined to form one coordinated system for the next control period. The 
desirable cycle length is calculated for the critical intersection in a group. This occurs 
through a look-up table based on the minimum delay cycle length calculations by the 
local controllers. The phase split is then calculated for the critical intersection. This is 
done by the local controllers to achieve equal degrees of saturation (where detection is 
available) or look-up tables in the case of minor intersections with limited detection. 
For non-critical intersections within the group, the phase split is generally determined 
by the minimum phase length requirements of the non-coordinated phases. 
10 
The other main function of the Strategic Level control is to select a common cycle 
length, matching phase sequence, and offsets for various groups and to pass these 
control constraints to individual controllers. 
The controller is having the ability to fine tune splits and offsets on a cycle-by-cycle 
basis, and to determine desired cycle lengths. Controller software is to provide this 
functionality. 
In order to measure degree of saturation, loops at the stop line will need to be sized to 
provide a monotonic relationship between time space and speed. This will require a 
loop array 16 feet long [24], In simple terms, the time space between vehicles is 
relatively independent of the size of following vehicles, but directly related to the 
speed of a vehicle. Within the range of non-congested speeds encountered at traffic 
signals, a loop array of 16 feet length will give a reasonable estimate of the speed of 
traffic, and an accurate count. It therefore provides the raw data for the degree of 
saturation and volume calculations. 
2.5 RHODES - Real time Hierarchical Optimized Distributed and Effective 
System: 
RHODES is the newest of these four systems having being produced in the mid 
1990's at the University of Arizona at Tucson [5], RHODES: Architecture, algorithms 
and analysis by Pitu Mirchandani..,et al [5] gives a detailed description about the 
system. 
The system takes input from the surface street detectors, predicts the future traffic 
streams at various hierarchical levels of aggregation, both specially and temporally 
and outputs the signal control settings that respond to these predictions. The 
optimization criterion can be any that is provided by the jurisdiction using the system 
but must be based on traffic measures of effectiveness such as average delays, stops, 
throughput, etc. 
At the highest level of RHODES is a "dynamic network loading" model that captures 
the slow-varying characteristics of traffic. These characteristics pertain to the network 
geometry (available routes including road closures, construction, etc.) and the typical 
11 
route selection of travelers. Based on the slow-varying characteristics of the network 
traffic loads, estimates of the load on each particular link, in terms of vehicles per 
hour, can be calculated. The load estimates then allow RHODES to allocate "green 
time" for each different demand pattern and each phase (North-South through 
movement, North-South left turn, East-West left turn, and so on). These decisions are 
made at the middle level of the hierarchy, referred to as "network flow control". 
Traffic flow characteristics at this level are measured in terms of platoons of vehicles 
and their speeds. Given the approximate green times, the "intersection control" at the 
third level selects the appropriate phase change epochs based on observed and 
predicted arrivals of individual vehicles at each intersection. 
Essentially, at each level of the hierarchy there is an estimation/prediction component 
and a control component. 
There are three aspects of the RHODES philosophy that make it a viable and effective 
system to adaptively control traffic signals. First, it recognizes that recent 
technological advances in communication, control, and computation make it possible 
to move data quickly from the street to the computing processors, make processing of 
this data to algorithmically select optimal signal timings fast and allow the flexibility 
to implement through modern controllers a wide variety of control strategies. Second, 
RHODES recognizes that there are natural stochastic variations in the traffic flow and 
therefore one must expect the data to stochastically vary. And third, RHODES 
proactively responds to these variations by explicitly predicting individual vehicle 
arrivals, platoon arrivals and traffic flow rates for the three corresponding levels of 
hierarchies. 
The two important issues to predict the traffic flow are the length of the time horizon 
and the number of prediction points per time horizon, which is called the prediction 
frequency. 
The prediction time horizon provides the real time traffic adaptive signal timing 
control logic with the ability to plan future signal timing decisions. If the prediction 
time horizon is short, perhaps several seconds, then the signal timing decisions are 
restricted. If the predictions are more over a 10 seconds time horizon, the signal 
timing logic can only make timing decision that extend or shorten the current phase. 
On the other hand, if the predictions are made over a long horizon, the signal timing 
12 
decisions can include decisions on phase sequencing and phase durations. The 
prediction frequency provides information about the distribution of vehicle arrivals 
over time. If the predictions are made at a frequency of only one prediction for the 
decision time horizon, then the signal timing logic must assume that the vehicles 
arrive uniformly during that time period. If the predictions are made more frequently, 
like every seconds over the prediction horizon, then the signal timing logic will have a 
more accurate representation of the distribution of vehicle arrivals over time. 
RHODES is a hierarchical control system [5] that uses predictive optimization, 
allowing intersection network levels of control. RHODES includes a main controller, 
a platoon simulator (APRES-NET), a section optimizer (REALBAND), an individual 
vehicle simulator (PREDICT) and a local optimizer (COP). The detector requirements 
for RHODES are fairly flexible. At minimum, RHODES requires upstream detectors 
for each approach to the intersections in the network. RHODES also can use stop-bar 
detectors to calibrate saturation flow rates and improve traffic queue estimates. 
The PREDICT algorithm in RHODES uses the output of the detectors on the approach 
of each upstream intersection, together with information on the traffic state and 
planned phase timings for the upstream signals, to predict future arrivals at the 
intersection. This approach allows a longer prediction time horizon since the travel 
distance to the intersection is longer and the delays at the upstream signal are 
considered. A benefit of this approach is that it includes the effects of the upstream 
traffic signals in the intersection control optimization problem. 
The several parameters are needed to be provided to the PREDICT model. They are 
the travel times on links ( detector to detector ) which depends on the link free flow 
speed and current traffic volumes, the queue discharge rates which also depends on 
volumes such as queue spillbacks and opposing and cross traffic volumes and the 
turning probabilities. 
Through-traffic queue discharge rates are effected by downstream through-traffic 
volumes, which can be easily measured. Likewise, left-turn queue discharge rates 
depend on opposing traffic volumes, and right-turn queue discharge rates depend on 
cross-traffic in that direction. These three discharge rates are initially given from 
13 
calculated default functions - functions of traffic volumes, but are then adjusted based 
on how well they predict remaining queues at the stop-bar presence detectors. 
The resolution of traffic at the network flow control level (i.e. level 2 of the RHODES 
hierarchy) is in platoons. Typically, RHODES will use a 20 - 40 second rolling 
horizon to predict arrivals and queues at each intersection, based on upstream detector 
data; at the network flow control level, RHODES will use a 200 - 300 second rolling 
horizon. 
Fixed control strategies are based on a signal-timing plan defined in terms of operating 
parameters for traditional signal control, namely cycle time, splits, and offsets. These 
parameters are generally developed based on traffic studies and standard procedures, 
such as the Highway Capacity Manual, or signal timing software such as TRANSYT 
and PASSER. The traffic studies result in estimates of traffic conditions, link volumes 
and turning percentages, for specified time periods. Signal timing parameters are 
developed for each of these time periods and, typically, implemented on a time-of-day 
basis with no consideration of current actual traffic conditions. In many cases, even 
the use of standard procedures for the development of signal timing plans is 
abandoned and traffic engineers operate in a judgment-based fashion with moderate 
levels of success. None of these approaches is truly traffic-adaptive or even attempt to 
actually minimize some measure of traffic performance such as average vehicle delay. 
The RHODES approach is to predict both the short-term and the medium term 
fluctuations of the traffic (in terms of individual vehicle arrivals and platoon 
movements respectively), and explicitly set phases that maximize a given traffic 
performance measure. RHODES does not necessarily require a prespecified phase 
sequence, but since many traffic engineers prefer a pre-specified sequence, RHODES 
has been developed to allow the traffic engineer to specify a desired sequence. In other 
word, in the RHODES control strategy, the emphasis shifts from changing timing 
parameters in reacting to traffic conditions just observed to proactively setting phase 
durations for predicted traffic conditions. 
Each decision has an associated value based on a performance measure such as stops 
or delay. This value is determined by using the predicted vehicle arrivals, the current 
and prior decisions, and an imbedded traffic flow model that accounts for estimated 
14 
queues, startup lost time, queue discharge and arrivals, as well as other traffic 
dynamics that relate the decision to the performance measure. 
The Dynamic Programming is completed when each possible decision for each stage 
has been evaluated in a forward recursion. Then a backward recursion is used to 
determine the sequence of phases and phase durations that will result in the lowest 
value of the performance measure over the optimization horizon. 
The decision for the first stage of the optimization is implemented as the desired 
signal control. Just prior to the end of this first phase, the optimization problem is 
solved again in a rolling horizon approach. The sequence of phases in the second 
optimization begins with the current phase which allow for the phase to be terminated 
early or extended based on the re-evaluation with more recent observations and 
predictions. 
Platoons are defined from observed detector data as a flow density above a pre-
specified level for some length of time [21]. Each platoon is characterized in terms of 
size (number of vehicles) and speed. When two (or more) platoons are predicted to 
arrive at an intersection and they request opposing signal phases, a conflict is said to 
occur. A decision tree is built where each branch of the tree represents one possible 
resolution of a conflict. The decision tree developed is based on the predicted platoon 
movement over some predefined horizon, such as 200-300 seconds, with node and 
two out-links for each conflict resolution. REALBAND evaluates, using APRES-
NET, the performance for each branch of the decision tree. When all branches have 
been explored, a path on the tree (corresponding to a set of conflict resolutions) is 
chosen with best estimated performance. 
The article, Manage traffic via innovative signal control in the web page of 
http: //ops. fhwa. dot. go v/publications/adaptivecontr ol/ [25] gives results oriented 
analysis of RHODES and OP AC. 
2.6 OP AC - Optimized Poicies for Adaptive Control: 
The central goal of signal control is to optimize vehicle flow through a given road 
network. In most cases, the most powerful and cost-effective approach is to improve 
the timing of the traffic signals at all intersections in the network. Adaptive signal 
15 
control uses very recent data to adjust signal timing "on the fly" based on close to 
real-time traffic conditions. Software systems like MIST (Management Information 
System for Transportation) and OP AC (Optimized Policies for Adaptive Control) 
have been used successfully by municipalities, state DOTs, and others for this 
purpose. OP AC is a system first proposed by Nathan Gartner at the University of 
Massachusets at Lowell in the early 1980's [25, 30], 
OPAC is a traffic adaptive signal control algorithm that provides for continuous online 
optimization of phase duration, cycle length, and offsets in immediate response to 
real-time traffic flows. OPAC eliminates the need to develop and store timing plans 
for different traffic conditions while controlling both isolated intersections and 
networks of intersections in coordinated operation. 
OPAC uses a predictive type of optimization with a rolling horizon [25]. This 
congestion control strategy—which attempts to maximize throughput—adjusts splits, 
offsets, and cycle length but maintains the specified phase order. For uncongested 
networks, OPAC uses a local level of control at the intersection to determine the phase 
online and a network level of control for synchronization, which is provided either by 
fixed-time plans (obtained offline), or a virtual cycle (determined online). 
The types of control and levels of local and global influence are flexible. Predictions 
are based on detectors located approximately 10 to 15 seconds upstream. After the 
initial 10 to 15 seconds, a model predicts traffic patterns. System monitoring and 
coordinated control features are provided through MIST, which communicates with 
individual intersection controllers and monitoring devices, gathering data to feed the 
OPAC algorithm and transmitting signal timing instructions rendered by OPAC. 
The implementation of an adaptive control system such as OPAC provides many 
advantages over existing fixed-timed plans by minimizing traffic delays through 
effective signal operation, improving travel time, reducing fuel consumption, and 
improving air quality while simultaneously reducing maintenance and operation costs 
and facilitating automated traffic data collection. 
Several U.S. cities, including Chicago and Seattle, already have such sophisticated 
computer-guided traffic management systems in place; Cary, N.C., and Sarasota, Fla., 
16 
are among those currently implementing such systems; and other cities around the 
nation are following suit [25, 18]. 
SCOOT and SCATS generally use a cycle based approach on a network, adjusting the 
cycle time, split and offsets among cycles in the network to optimize a measure of 
effectiveness [3, 18]. OP AC and RHODES vary somewhat from this, with OPAC 
being cycle based and RHODES optimizes a measure of effectiveness over a fixed 
prediction horizon and then extends the horizon by a fixed time step and reiterates the 
optimization until an optimal split of the given cycle is found. Advancements in 
approaches to OPAC have allowed for some variability in network wide cycle lengths 
[18]. 
The main objective of the research is to make a simulation model of an adaptive traffic 
control system, which is better to implement in Colombo city to minimize delay 
caused by existing fixed time traffic control system. The detection method for the 
proposed system and the design methodologies are described by the next chapter. 
17 
Chapter 3: Decentralized Intelligent Control Model 
3.1 Introduction: 
The essential feature of the real time traffic control system is its capability to identify 
the real time traffic load on the road. And the key factor to be sensed in the control 
strategy is the influence to the approach. After getting the vehicle counts from each 
detector, the influence is to be known to decide the respective split or green phase. In 
this chapter, function of the detection, the surrounding traffic data collection methods 
and the inductive loop detectors, which are selected for this research, are discussed. 
Later, the way of getting influence and the effect of detection values to the influence 
are discussed. 
3.2 Detection in adaptive Control: 
3.2.1Function of detection: 
In the adaptive traffic control process in its current state, field detection serves to 
obtain the arrival times for vehicles approaching an intersection. In the field, this 
arrival time is difficult to measure. In the simplest case one detector is placed and the 
placement it installed gives its own issues [9]. 
When the detector is installed at the stop line, the measurement of the arrival time at 
the intersection of vehicles moving through a green phase, but becomes problematic 
for vehicles arriving during a stop phase, since only a small number of vehicles can sit 
on a detector and have their arrival times accurately recorded. Vehicles that proceed 
after the red signal has turned green will not have their desired arrival times recorded. 
When the detector is installed at a distance upstream of the stop line, allows 
determination of the arrival time of a vehicle continuously moving at its desired speed 
at the detector and through the intersection, but may problematic for vehicles in other 
situations, generally those that are required to change the speed between the detector 
and the intersection, such as to stop when encountering a signal state change, to slow 
to make a turn, to avoid an incident or to perform a passing maneuver. 
18 
V* u ! 
wan ws mwtimmwmwt 
AMWWKIa 
The preferable approach of the two is to place the detector upstream of the stop line as 
this will allow at least some determination of the arrival time of a vehicle in any 
situation. 
3.2.2 Detection methods: 
Detection technologies should involve the following factors [3, 4, 14], 
•Cost: Utilization of as much of the existing detector equipment as possible should be 
considered. If the replacement of that equipment required, it should be low cost in 
terms of both installation and maintenance. 
•Practicality: Consideration should include the ability to place the desired detectors 
where they are needed, or whether issues such as proximity of intersections in a 
network, weather conditions or physical obstacles may hinder placement. Also to be 
considered is the conformity of the detector system in a network that is detector types 
and configuration should be as uniform as possible throughout a network. 
•Accuracy: The ability of the detector technology to provide data that reflect actual 
field conditions is vital. Sensitivity, ability to deal with various conditions and 
deterioration of data accuracy over time are issues to reflect. 
The most common detector technology in world is the inductive loop detector, in 
which generates a current as a vehicle passes over it due to the vehicle's magnetic 
influence on the inductive loop [29], 
Advanced detection technology may allow, easier implementation of the proposed 
detection strategies. This is especially true in the case of multi-lane urban network 
applications where incidents erratic driver behaviour and transit use make the ability 
to identify the specific vehicles important to gathering data. Video detection is such a 
technology. But video detection has some drawbacks. That is physical placement may 
be difficult because of availability of mounting structures, blocking by obstacles and 
angles required for proper detection, and they may also not operate as effectively at 
night or during obscuring weather conditions. 
91207 19 
3.2.3 Inductive loop detectors: 
Considering the main advantages of cost effectiveness and the reliability, Inductive 
loop detectors have been selected as the detection method. 
The inductive loop consists of a multi-winding wire coil installed in the floor and 
makes up the actual sensor element. The system for the contact-free detection of 
metallic objects is formed by the inductive loop and the switching unit, referred to as 
loop [2], 
The inductive loop and a capacitor which is integrated in the loop detector form an LC 
oscillator. The frequency of resonance of this resonant circuit is determined by the 
capacitance of the capacitor and the magnitude of the loop inductance. The 
capacitance of the capacitor and thus the resonance frequency can be modified using 
an internal switch. This prevents interference between two adjacent inductive loops or 
detectors (Figure. 3.1). 
Figure 3.1-Block diagram of the Inductive Loop 
The general rule is that the higher the inductance, the lower the frequency of 
resonance. 
If a current is present in the loop, an alternating magnetic field is generated around it. 
The magnetic field lines get closed to form loops. When a vehicle moves onto the 
inductive loop, currents are induced by the alternating field of the loop in the metallic 
parts of the vehicle (chassis, axles). A magnetic field is generated by these currents, 
20 
which opposes the loop field. This results in a change in the loop inductance and thus 
the frequency. This change in frequency is detected by the loop detector. 
Depending on the detector type, the output relay is either energized or released when a 
predefined switching threshold is exceeded. This switching threshold can be adjusted 
using the sensitivity switch (High - Medium - Low). 
«High» is meant when the switching threshold is low and «Low» is meant when the 
threshold is high. The sensitivity should always be adapted to the respective area of 
application. The output relay is not switched on by slow frequency changes (e.g. due 
to temperature variations). 
In general, implementation of the inductive loop detectors is to enhance detection 
process, which is an essential thing in influence calculation to determine the optimum 
split. 
3.3 Influence Detection: 
3.3.1 Structure of the intelligent control: 
The intelligent control mechanism can be defined under a number of models. They are 
simply the detection model as discussed earlier, the prediction model and the Decision 
model as discussed below [3, 11, 12], 
•Detection: This is the process by which vehicles that enter a given approach to an 
intersection are recognized for processing by the adaptive control scheme. In this 
research, inductive loop detectors are recommended to use considering the cost 
effectiveness and the reliability in use. 
•Prediction: This is the process by which the data from the detection process are used 
to determine the influence to the respective approach at the intersection. This has to be 
done considering the vehicle counts over the cycle time to predict the influence to the 
given approach in next cycle. 
21 
•Decision: This is the process by which the predicted influences are used to decide 
green times to the various approaches of the intersection. In this research, this green 
time is decided by a Fuzzy inference system. 
3.3.2 Formulating the Influence function for traffic signal control: 
The prediction model is for the purpose of forecasting possible traffic flow in a 
particular intersection. As shown in figure 3.2, it is noticed that the traffic flow of 
intersection A is influenced by vehicles passing through nearby intersections. The 
influence of nearby intersections to intersection A is inversely proportional to the 
distance between these intersections [1], In other words, the far the two intersections 
are, the less the influence is. 
u 1 N1 • • • N2 M t t M N 3 
J I • rrr 
Figure 3.2- part of the traffic network 
As same as the distance, the number of vehicles enter to the particular approach also 
add a considerable effect to the influence. It is noticed that the vehicle count is direct 
proportional to the influence. Hence the influence can be expressed in the following 
way [1], 
The parameter INFLUENCE a No. of vehicles entering the phase 
a 1/distance 
INFLUENCE = 1/distance * car queue. (1) 
According to the equation (1), the influence is calculated and be input to the fuzzy 
inference system to decide the green phase or split. The factor influence for fuzzy 
22 
inference system stands in between the range of zero to the maximum of one. Hence 
for each intersection, it has to decide separately for all approaches and taken as a 
factor to the maximum influence of the same intersection. 
The considered intersection A as shown in figure 3.2 is detailed and shown in figure 
3.3. As shown there, the considered intersection is 300 m from West, 1 Km from 
North, 500 m from East and 1 Km from South neighboring intersections. Inductive 
Loop detectors are to be installed as shown. The relay outputs of each detector are the 
inputs to the intersection controller. 
B 
A 
•! • 
o 
• u 
• SCI 
• SC2 
c 
D 
Figure 3.3-Intersection A 
SA1 and SA2 detectors are to detect the vehicles entering to the West Approach A. 
Likewise Sm, SB2 are to detect the vehicles entering North Approach B, Sn , SC2 are 
to detect East Approach C count and Sm, SD2 are to detect the South Approach D 
vehicle counts. The total count will reach the intersection at the average speed of 50 
Km/h. If the average vehicle length is 3 m, the maximum number of vehicles which 
can pass across the intersection is experimented as 275 vehicles during 60 sees time 
23 
duration when one approach is given its green phase. The maximum influence is given 
when the vehicle count exceeds 275 in West approach A, which is having the lowest 
distance to the mentioned intersection. Hence the calculated influence is taken as a 
factor, divided by this maximum influence. 
The following assumptions were made for the formulation of each approach influence. 
• 85% of phase vehicle count at the intersection is to enter to the front phase 
• 15% of Left track vehicle count is to left turn 
• 15% of Right track vehicle count is to Right turn 
Basic formula for the influence for the particular approach is given by the equation (2) 
INFLUENCE = (0.85 x direct phase Influence) + (0.15 x Left phase Influence) 
+ (0.15 x Right phase Influence) (2) 
Hence the Influence at each approach is as follows w h e r e d A , d B , dc and dD stand 
for the distance from neighboring intersections starting from west to south to the 
particular intersection. 
• INFLUENCE A = l/dc x0.85 (5C1 +SC2)+\/dB x0.15 Sm + 
VdDx0.\5SD2 (3) 
• INFLUENCE B = l/dD x0.85 (Sm +SD2)+ \/dA x0.15 SA2 + 
\ldc x0.15 Scl (4) 
• INFLUENCE C = MdA x0.85 (SM +SA2) + \/dD x0.15 Sm + 
l/dB x 0.15 SB2 (5) 
• INFLUENCE D = \/dH x0.85 (Sm + SB2) + \/dA x0.15 SM + 
\ldc x 0.15 SC2 (6) 
Getting those influences into the account, the green phase for each approach 
decided by the Fuzzy Inference System as discussed by the next chapter. 
is 
24 
Chapter 4: Fuzzy Control Strategy Development 
4.1 Introduction: 
Getting the influence described in last chapter for the input of the Fuzzy Inference 
System, the Split is decided. Fuzzy sets of the input and the output, Fuzzy rule 
development and the way of deciding the split according to the changing traffic load 
are described in this chapter. Detectors give their values periodically. Getting those 
values in to account, the influence is calculated by the intersection controller and those 
influence values are taken for the input to the fuzzy inference system. FIS determines 
the green phase for each approach. Starting from detector value count to the output 
phase light energizing, the controller has to act as the brain of the intersection control 
system. The way of control mechanism developed and the way of controller behave 
accordingly are discussed in this chapter. 
4.2 Fuzzy Inference System 
4.2.1 Fuzzy basics: 
The concept of Fuzzy Logic (FL) was conceived by Lotfi Zadeh, a professor at the 
University of California at Berkley, and presented not as a control methodology, but 
as a way of processing data by allowing partial set membership rather than crisp set 
membership or non-membership. This approach to set theory was not applied to 
control systems until the 70's due to insufficient small-computer capability prior to 
that time [8], 
Fuzzy Logic incorporates a simple, rule-based IF X AND Y THEN Z approach to a 
solving control problem rather than attempting to model a system mathematically. The 
Fuzzy Logic model is empirically-based, relying on an operator's experience rather 
than their technical understanding of the system. 
Fuzzy Logic requires some numerical parameters in order to operate such as what is 
considered significant error and significant rate-of-change-of-error, but exact values of 
these numbers are usually not critical unless very responsive performance is required 
in which case empirical tuning would determine them. 
Fuzzy Logic was conceived as a better method for sorting and handling data but has 
proven to be an excellent choice for many control system applications since it mimics 
25 
human control logic. It can be built into anything from small, hand-held products to 
large computerized process control systems. It uses an imprecise but very descriptive 
language to deal with input data more like a human operator. It is very robust and 
forgiving of operator and data input and often works when first implemented with 
little or no tuning. Fuzzy inference systems have been successfully applied in fields 
such as automatic control, data classification, decision analysis, expert systems, and 
computer vision. Because of its multidisciplinary nature, fuzzy inference systems are 
associated with a number of names, such as fuzzy-rule-based systems, fuzzy expert 
systems, fuzzy modeling, fuzzy associative memory, fuzzy logic controllers, and 
simply (and ambiguously) fuzzy systems. The design of a fuzzy inference system 
includes fuzzification of inputs and outputs with a knowledge base that contains the 
membership functions and the rule sets and the defuzification method to get the 
reasoned output values. In this research, a single input single output fuzzy inference 
system is used to decide the split for each approach. Four fuzzy inference systems are 
to be implemented in one intersection controller to decide each green phase for four 
approaches at the intersection. 
4.2.2 Fuzzification: 
Mamdani's fuzzy inference method [22] is the most commonly seen fuzzy 
methodology. Mamdani's method was among the first control systems built using 
fuzzy set theory. The fuzzification comprises the process of transforming crisp values 
into grades of membership for linguistic terns of fuzzy sets. The membership function 
is used to associate a grade to each linguistic term. 
Influence is taken as the fuzzy input. Since it has been set as a factor compared to the 
maximum possible influence, it gets a value in between zero and one. Fuzzification is 
the process by which system input is evaluated to determine the degree at which it 
belong to a particular fuzzy set, on a scale from 0 to 1. The system input has been 
categorized into five fuzzy sets as mentioned in Table 4.1. 
A membership function (MF) is a curve that defines how each point in the input space 
is mapped to a membership value (or degree of membership) between 0 and 1. The 
input space is sometimes referred to as the universe of discourse, a fancy name for a 
simple concept. 
26 
If X is the universe of discourse which is the influence which stands in between 0 to 1 
and its elements are denoted by x, then a fuzzy set A in X is defined as a set of 
ordered pairs [22], 
A = {x, nA(x) | x e X} 
The membership function nA(x) of the input influence categorization is taken as 
trapezoidal membership function, which has a flat top and really is just a truncated 
triangular curve. These straight line membership functions have the advantage of 
simplicity [8]. 
A M * ) 
Very Low 0 - 0 . 2 
Low 0 - 0 . 6 
Average 0 .2 -0 .75 
High 0 . 5 - 1 
Very High 0.8- 1 
Table 4.1- Membership values of the Input Influence 
In Figure 4.1, the membership functions with their values are graphically represented. 
The y-axis in Figure 4.1 represents the degree of truth from completely false (0) to 
completely true (1). The x-axis represents the range of input values for the particular 
system input. 
Average High 
input variable "Influence" 
Figure 4.1- Trapezoidal Membership Function for the Input Influence 
27 
The membership function |iA(x) of the output Split categorization is taken 
triangular membership function, which has a sharp top. 
A jiA(x) (Sec.s) 
Very short 0 - 9 
Short 1 0 - 2 0 
Average 2 5 - 3 5 
Long 4 5 - 5 5 
Very long 5 5 - 6 0 
Table 4.2- Membership values of the Output Split 
In Figure 4.2, the membership functions with their values for the output Split are 
graphically represented. The y-axis in Figure 4.2 represents the degree of truth from 
completely false (0) to completely true (1). The x-axis represents the range of output 
values for the particular system output 0 to the maximum of 60 sees.. 
Very^ort S h o r t A y e r a g e 
output variable "Split" 
Figure 4.2 -Triangular Membership Function for the Output Split 
The membership function is a graphical representation of the magnitude of 
participation of each input [8], It associates a weighting with each of the inputs that 
28 
are processed, define functional overlap between inputs, and ultimately determines an 
output response. The rules use the input membership values as weighting factors to 
determine their influence on the fuzzy output sets of the final output conclusion. Once 
the functions are inferred, scaled, and combined, they are defuzzified into a crisp 
output which drives the system. There are different membership functions associated 
with each input and output response. The membership functions associate a weighting 
factor with values of each input and the effective rules. These weighting factors 
determine the degree of influence or degree of membership (DOM) each active rule 
has. By computing the logical product of the membership weights for each active rule, 
a set of fuzzy output response magnitudes are produced. All that remains is to 
combine and defuzzify these output responses. 
Figure 4.3 represents the variation of Split according to the given influence. It can be 
noticed that with the increase of influence to the particular approach, the Split also 
increases accordingly. 
Figure 4.3 -Variation of Split Vs. Influence 
29 
4.2.3 Rule base: 
As inputs are received by the system, the rulebase is evaluated. The antecedent (IF X 
AND Y) blocks test the inputs and produce conclusions. The consequent (THEN Z) 
blocks of some rules are satisfied while others are not. The conclusions are combined 
to form logical sums. These conclusions feed into the inference process where each 
response output member function's firing strength (0 to 1) is determined. A firing 
strength for each output membership function is computed. All that remains is to 
combine these logical sums in a defuzzification process to produce the crisp output. 
Fuzzy sets and fuzzy operators are the subjects and verbs of fuzzy logic. These if-then 
rule statements are used to formulate the conditional statements that comprise fuzzy 
logic. A single fuzzy if-then rule assumes the form [22] 
if x is A then y is B 
where A and B are linguistic values defined by fuzzy sets on the ranges (universes of 
discourse) X and Y, respectively. The if-part of the rule "x is A" is called the 
antecedent or premise, while the then-part of the rule "y is B" is called the consequent 
or conclusion. 
In this research single input single output Fuzzy Inference System is used and hence 
simple five rules were developed. Therefore complex operators are not required to get 
the output value determined. Those rules are mentioned below. 
If influence is VERY LOW, then split is VERY SHORT 
If influence is LOW, then split is SHORT 
If influence is AVERAGE, then split is AVERAGE 
If influence is HIGH, then split is LONG 
If influence is VERY HIGH, then split is VERY LONG 
Rule evaluation is how Fuzzy Logic performs calculations. The fuzzy values produced 
membership functions are passed through the rule list to find the fuzzy output. The 
consequent specifies a fuzzy set be assigned to the output. The implication function 
then modifies that fuzzy set to the degree specified by the antecedent. The most 
common ways to modify the output fuzzy set are truncation using the min function 
30 
(where the fuzzy set is "chopped off ' ) or scaling using the prod function (where the 
output fuzzy set is "squashed"). Both are supported by the Fuzzy Logic Toolbox. 
Figure 4.4 represents the graphical representation of rule evaluation to a particular 
input value. 
\ 
Figure 4.4- Graphical representation of rule evaluation. 
In Figure 4.4, the way of split determination to the input value of the influence at 0.5 
is shown. When the influence is at 0.5, it gives a membership value of zero to the 
membership function VERY LOW. In the same way, it gets 0.45 of membership value 
from the membership function LOW, the maximum of "completely true" membership 
value 1 from the membership function AVERAGE and the membership value 0 from 
both membership functions HIGH and VERY HIGH. That grading is shown at the left 
hand side of the diagram. At the right hand side of the diagram, the split determination 
is shown according to each rule. At the bottom of the right hand side, it is shown the 
final determination of the split according to the averaging technique. After getting 
each split with reference to all of the rules mentioned above, they are averaged to 
determine the final best fit value for the output split. In this example it was determined 
as 24.3 seconds to be the split for 0.5 of the influence. 
31 
Interpreting if-then rules is a three-part process. This process is explained in detail 
below [8, 22], 
1. Fuzzify inputs: Resolve all fuzzy statements in the antecedent to a degree of 
membership between 0 and 1. If there is only one part to the antecedent, this is 
the degree of support for the rule. 
2. Apply fuzzy operator to multiple part antecedents: If there are multiple parts to 
the antecedent, apply fuzzy logic operators and resolve the antecedent to a 
single number between 0 and 1. This is the degree of support for the rule. 
3. Apply \implication method: Use the degree of support for the entire rule to 
shape the output fuzzy set. The consequent of a fuzzy rule assigns an entire 
fuzzy set to the output. This fuzzy set is represented by a membership function 
that is chosen to indicate the qualities of the consequent. If the antecedent is 
only partially true, (i.e., is assigned a value less than 1), then the output fuzzy 
set is truncated according to the implication method. 
The output of each rule is a fuzzy set. The output fuzzy sets for each rule are then 
aggregated into a single output fuzzy set. The logical product of each rule is inferred 
to arrive at a combined magnitude for each output membership function. This can be 
done by max-min, max-dot, averaging, RSS, or other methods. Once inferred, the 
magnitudes are mapped into their respective output membership functions, delineating 
all or part of them. The "fuzzy cento id" of the composite area of the member 
functions is computed and the final result taken as the crisp output. Tuning the system 
amounts to "tweaking" the rules and membership function definition parameters to 
achieve acceptable system response. Finally the resulting set is defuzzified, or 
resolved to a single number. 
Since decisions are based on the testing of all of the rules in an FIS, the rules must be 
combined in some manner in order to make a decision. Aggregation is the process by 
which the fuzzy sets that represent the outputs of each rule are combined into a single 
fuzzy set. Aggregation only occurs once for each output variable, just prior to the final 
step, defuzzification. The input of the aggregation process is the list of truncated 
output functions returned by the implication process for each rule. The output of the 
aggregation process is one fuzzy set for each output variable. 
32 
4.2.4 Determination of the single output: 
The final step in the Fuzzy Logic calculation is defuzzification, when the raw fuzzy 
outputs are evaluated to create a composite system output. 
Mamdani-type inference, which has been defined in the Fuzzy Logic Toolbox, expects 
the output membership functions to be fuzzy sets. After the aggregation process, there 
is a fuzzy set for each output variable that needs defuzzification. It's possible, and in 
many cases much more efficient, to use a single spike as the output membership 
function rather than a distributed fuzzy set. Hence a triangular shaped spike has been 
used as the membership function for the split as shown in Figure 4.2. This is 
sometimes known as a singleton output membership function, and it can be thought of 
as a pre-defuzzified fuzzy set. It enhances the efficiency of the defuzzification process 
because it greatly simplifies the computation required by the more general Mamdani 
method, which finds the centroid of a two-dimensional function. Rather than 
integrating across the two-dimensional function to find the centroid, we use the 
weighted average of a few data points. 
In the Fuzzy Logic Toolbox, there are five parts of the fuzzy inference process: 
fuzzification of the input variables, application of the fuzzy operator (AND or OR) in 
the antecedent, implication from the antecedent to the consequent, aggregation of the 
consequents across the rules, and defuzzification. 
Once proper weighting has been assigned to each rule, the implication method is 
implemented. A consequent is a fuzzy set represented by a membership function, 
which weights appropriately the linguistic characteristics that are attributed to it. The 
consequent is reshaped using a function associated with the antecedent (a single 
number). The input for the implication process is a single number given by the 
antecedent, and the output is a fuzzy set. Implication is implemented for each rule. 
The input for the defuzzification process is a fuzzy set (the aggregate output fuzzy set) 
and the output is a single number. As much as fuzziness helps the rule evaluation 
during the intermediate steps, the final desired output for each variable is generally a 
single number. However, the aggregate of a fuzzy set encompasses a range of output 
values, and so must be defuzzified in order to resolve a single output value from the 
33 
set. Perhaps the most popular defuzzification method is the centroid calculation, 
which returns the center of area under the curve. 
As shown in Figure 4.2, the split for the influence at 0.5 is determined in this way and 
the single defuzzified output split has been determined as 24.3 seconds. 
Four Fuzzy Inference Systems are implemented to decide each four splits at the 
intersection and the control strategy to control the changing traffic load at real time is 
discussed below. 
4.3 Control Strategy 
4.3.1 System Architecture: 
The basic structure of the proposed system includes one independent controller for 
each intersection. The functions that a system intersection controller performs are the 
following. 
• It reads the detector's state and hence it computes traffic flow information. 
• It computes the timing schedule for the intersection based on traffic flow. 
• It switches traffic lights on and off 
The whole system can be integrated in a programmable logic controller or in a micro 
controller. The number of inputs should be equal to the number of detectors and the 
number of outputs should be the number of light signals connected to that particular 
approaches. For the intersection shown in Figure 3.3, the system should include eight 
digital inputs module completes with twenty four outputs module integrated to the 
controller. 
Figure 4.5 denotes the block diagram of the control mechanism. Getting influence and 
the delayed cycle time as system inputs, the rule base determines the values for the 
approach split for each rule. They are aggregated and defuzzified to get a single value 
for the split. By totalizing four splits, the cycle time is determined. 
34 
Figure 4.5- Block diagram of the Control System 
4.3.2 Decentralized approach: 
Many adaptive traffic control systems developed in the world are centralized, network 
control systems. There one intersection controller is placed at the intersection and 
performs all functions and computations required at the intersection. The intersection 
controllers are connected to a central computer named master controller. Intersection 
controllers act according to the instructions given by the master controller and the 
master controller calculates the split, offset and cycle time required to the control of 
traffic load. Hence the total traffic grid is performed for a common cycle. Figure 4.6 
shows the way of the communication network built in many other traffic control 
structures, which are having a network control approach [2], 
35 
Figure 4.6-Master Slave Control Model 
In the proposed system, a decentralized method is introduced. Each intersection at the 
grid is having its own intersection controller and this controller calculates all the 
parameters such as split, cycle and offset required to control the traffic load at that 
intersection. Hence a master controller is not needed. As a result of this independency, 
intersections at the grid do not perform according to a common optimizing method 
and hence do not having common parameters at all. With decentralized control 
strategy, controller may be able to react to incident with short response time and in a 
proactive way. Since no master controller is required and not functioning according to 
a common algorithm for the whole grid, adding new intersections to the network 
increases only the computation loading on neighborhood intersections while rest of 
the network remains same. 
4.3.3 Prediction method: 
For proactive traffic control, it is important to predict vehicle arrivals, turning 
probabilities and queues at intersections, in order to compute phase timings that 
optimize a given measure of effectiveness [6, 23]. Considering the intersection shown 
in Figure 3.3, it has four approaches. Associated with each approach are several 
possible traffic movements of left turn, right turn and a through movement. Any non 
36 
conflicting combination of movements that can share the intersection at any one time 
can be assigned a single phase that allows those movements protected use of the 
intersection. 
In the proposed system, detectors are to be installed just after the neighboring 
intersections hence vehicles are get counted at the beginning of the path to the 
particular intersection. Thus, the vehicles will get a chance to be on travel after got 
counted by the relevant detector and the chance has been maximized to have the green 
phase get activated for the approach to enter, when they reach the intersection. As a 
result of that, the delay has been minimized. 
The vehicle flow pattern is identical to each intersection. If both cross roads have 
equal priorities as shown in Figure 3.3, High priority is given to the through 
movement and 15% of vehicle count is assumed to get left or right turns. But this 
should be differing from intersection to intersection. If one road gets more priority 
than the other crossed road as if it is a crossed lane, high priority is given to the main 
road and these turning vehicle percentages are to be got after doing an analysis of 
these vehicles movement on this intersection along peak hours. 
Two issues are important to predicting traffic flow. They are the length of the time 
horizon and the number of prediction points per time horizon, which is the prediction 
frequency [3], The prediction time horizon provides the real time traffic adaptive 
signal timing control logic with the ability to plan the future signal timing decisions. 
It is assumed as the traffic load at the current cycle time is similar to the load got in 
the previous cycle. If differences are there, it gives negligible effect since the longest 
cycle time allowed for all intersections .is 240 seconds. Hence considerable change 
cannot be predicted in adjacent time periods. Hence it uses the output of the detectors 
on the approach as inputs of the controller and inbuilt counters get the counted 
numerical values of the moved vehicles across the detectors. These counted values are 
taken for the next phase and cycle time prediction. The signal timing logic assumes 
that the vehicles arrive uniformly during the particular cycle time and were taken for 
next cycle calculations. 
4.3.4 Control mechanism: 
Figure 4.7 shows the control flow diagram of the control strategy. The system adapts 
its timing to traffic demands by continuously changing the cycle time. Each cycle may 
37 
have a different duration compared to the previous cycles. The cycle time is computed 
by the intersection controllers in a distributed fashion at the end of each cycle. The 
computation of that cycle time is based on traffic measurements taken during the 
previous cycle. The cycle time can change according to the traffic demand. 
Before the beginning of a traffic light cycle, the controller computes the influence 
according to the detector measurements during the previous cycle. From these 
influences, the optimal green phases for each approach are computed as described in 
chapter 3. 
At the initial start up, all the splits of all approaches are taken as 15 seconds and the 
cycle time is to be 60 seconds. The output timers are set in to these values and the 
light signals are energized accordingly. Starting from approach A, green phase is 
given 15 seconds each and rotates clockwise till approach D gets its green phase on. 
While the cycle is running, detectors energize their relay outputs for each vehicle 
passing across. These output values are the inputs to the controller and the counters in 
control program count these input pulses. Influence is calculated by the controller and 
the phase values are determined. By totalizing those four splits, cycle time is 
computed. After the determination of those splits, the output timers are given those 
split values and the counters are reset. Then the next cycle run is started with 
energizing the traffic signals for the determined time period at its green phase. After 
that cycle run finished, the timers are reset and the next cycle run is started. The same 
methodology continues and repeats on all subsequent cycles till the controller gets 
power off or any kind of fault detection. 
In order for a controller to compute the optimal timing for the next cycle, it must 
predict traffic conditions. Thus, the controller assumes that during the next cycle the 
distribution of arrivals will be exactly the same with that of previous cycle. For that 
reason, the controller keeps detailed information for the arrivals during the complete 
cycle time. 
38 
Figure 4.7- Flow Chart for Proposed Adaptive Control Mechanism 
39 
The strategy for determining green to each approach and repeating the same sequence 
of control program has been developed as shown in Figure 4.7 and the simulation 
model with simulation results for a set of data and comparison results against the 
existing fixed time traffic control system are described in the next chapter. 
40 
Chapter 5: Simulation and Results 
5.1 Introduction: 
The performance of the mechanism is tested by simulation. Simulation model was 
developed by Matlab Simulink and the study area was the Narahenpita junction at 
morning peak hours. The details of the simulaton model and the simulation results are 
described in this chapter. Later the use of traffic data derived from the previous steps 
in the process to implement a traffic signal control scheme is examined. Three sets of 
data from Table 5.1 were taken to do the delay calculation. Both fixed time control 
system and the intelligent model are compared. 
5.2 Simulation Model Development 
5.2.1 Simulink basics: 
Simulink is a software package that enables to system model, simulate, and analyze 
systems whose outputs change over time. Such systems are often referred to as 
dynamic systems. Simulink can be used to explore the behavior of a wide range of 
real-world dynamic systems. Simulating a dynamic system refers to the process of 
computing a system's states and outputs over a span of time, using information 
provided by the system's model [22]. 
At the start of the simulation, the model specifies the initial states and outputs of the 
system to be simulated. At each step, Simulink computes new values for the system's 
inputs, states, and outputs and updates the model to reflect the computed values. At 
the end of the simulation, the model reflects the final values of the system's inputs, 
states, and outputs. 
A Simulink Block Diagram is a pictorial model of a dynamic system [22], It consists 
of a set of symbols, called blocks, interconnected by lines. Each block represents an 
elementary dynamic system that produces an output either continuously (a continuous 
block) or at specific points in time (a discrete block). The lines represent connections 
of block inputs to block outputs. Every block in a block diagram is an instance of a 
specific type of block. The type of the block determines the relationship between a 
41 
block's outputs and its inputs, states, and time. A block diagram can contain 
number of instances of any type of block needed to model a system. 
5.2.2 Traffic simulation model of an intersection: 
It is clear that any type of traffic control algorithm needs to be tested in the laboratory 
before it is implemented and evaluated in the field. The most appropriate method to do 
this testing is to have a realistic simulation model of traffic flow at an intersection, 
emulate the traffic detections and observe the resulting changes that would come 
about if the algorithm was implemented in place of the current control system. Hence 
the simulation was done with the Matlab developed simulation model. 
Figure 5.1 shows the simulation model developed for the intersection shown in Figure 
3.3. In the left hand side, the eight detectors are shown. According to the equations (2) 
- (4), certain percentages of the detector values are taken to calculate the influence. 
Hence the summation blocks add the detectors and from the gain block, influence is 
calculated and given as a factorized output, which stands in between 0 and 1 to the 
Fuzzy Inference System with rule viewer. Four Fuzzy Inference Systems are shown 
for each separate approach. The influence parameters for each are separately 
developed prior to insert as a gain at the gain block. The right hand side display blocks 
provide the output of the Fuzzy Inference Systems and they are the green phases for 
each approach as shown. At the bottom display, it is shown the cycle time for those 
particular approach splits. 
42 
Figure 5.1 Traffic Simulation Model developed by Matlab 
43 
5.2.3 Simulation results: 
Contained in Table 5.1 are the associated statistics for set counts of generated vehicle 
arrivals to the intersection shown in Figure 3.3. Detector count values are shown there 
at five cycle times. 
Contained in Table 5.2 are the influences calculated and Table 5.3 gives the splits 
determined for the data, given to the fuzzy inference systems. 
A B C D CYCLE 
*A2 Sc. SC2 S D1 ® D2 
60 65 25 20 100 100 5 15 60 
103 95 34 30 160 135 16 20 62 41 
162 175 93 110 225 210 53 50 94.94 
190 200 118 123 230 243 75 72 151 1 
221 218 194 190 302 320 201 198 155.7 
90 92 35 40 85 81 44 52 190 3 
81 78 67 70 53 56 33 38 80 32 
52 55 75 78 34 38 22 25 67 32 
63 68 51 55 108 115 89 94 56 64 
20 26 18 22 15 12 8 5 76 22 
16 12 8 5 2 0 4 6 34 49 
112 118 102 98 89 94 98 92 25 83 
202 198 181 190 178 185 78 83 106 7 
120 115 32 38 4 6 51 58 149 7 
5 12 115 112 3 6 41 38 76.6 
Table 5.1- Detector counts at Approach Counters 
A B C D CYCLE 
0.198 0.06 0.077 0.105 62.41 
0.29 0.102 0.125 0.15 94.94 
0.448 0.244 0.24 0.425 151.1 
0.499 0.329 0.281 0.504 155.7 
0.702 0.808 0.371 0.783 190.3 
0.285 0.167 0.576 0.145 80.32 
0.227 0.125 0.508 0.189 67.32 
0.127 0.085 0.347 0.183 56.64 
0.177 0.242 0.428 0.17 76.22 
0.046 0.031 0.147 0.052 34.49 
0.046 0.016 0.088 0.021 25.83 
0.304 0.269 0.743 0.277 106.7 
0.567 0.316 0.99 0.515 149.7 
0.372 0.165 0.741 0.132 76.6 
0.014 0.081 0.077 0.215 41.37 
Table 5.2- Influences Calculated for the Detector Data 
44 
Table 
A B C D I CYCLE 
21.69 9.16 21.87 9.75 62.41 
32.54 10.98 39.05 12.37 94.94 
51.81 19.07 59.01 21.18 151 1 
53.99 20.71 58.93 22.09 155.7 
58.95 36.73 58.8 35.86 190 3 
20.74 13.02 35.18 11.38 80.32 
16.45 10.43 25.74 14.7 67.32 
11.68 9.05 21.21 14.71 56.64 
22.5 17.99 22.37 13.36 76.22 
8.06 6.93 11.5 7.99 34.49 
4.71 5.8 9.14 6.17 25.83 
21.59 19.27 46.26 19.55 106 7 
43.39 20.6 58.91 26.76 149.7 
7.06 12.84 45.98 10.73 76.6 
7.52 
C 1 O 1-j 
8.9 8.83 16.1 41.37 
Considering the above results, it can be noticed that when detector values getting high 
the influence also getting increased and hence the split determined by the FIS is also 
getting long. Thus, it can be noticed that the system is behaving according to the real 
time traffic load. A comparison done with existing fixed time control system and the 
improvements obtained by the implementation of the real time traffic control system 
are discussed below. 
45 
5.3 Comparison with existing Fixed time Traffic Control System 
5.3.1 Delay calculation: 
The delay calculation was done to compare the intelligent model with the existing 
fixed time traffic control system. 
In considering stopped delay, there are three possible conditions for which a vehicle 
can arrive at any given approach [3, 6]. These are: 
• A vehicle can arrive while the approach does not have a green phase, but before the 
time the green phase begins. 
• A vehicle can arrive while the approach has a green phase. 
• A vehicle can arrive while the approach does not have a green phase, and after the 
time the green phase has ended. 
If a vehicle arrives while the approach does not have a green phase in either condition, 
it must experience stopped delay, while it will experience no stopped delay if it arrives 
during the approach's green phase. 
Figure 5.2 shows the way of assigning splits of four approaches of the intersection 
shown in Figure 3.3 over the cycle time. Starting from approach A, splits are assigned 
for all four approaches one after the other. If a vehicle reaches the stop line at tl and is 
to enter to the approach A, it can pass the intersection without delay. But the vehicle 
came at the same time and is expecting to enter to the approach B, it has to wait till the 
split B of the same cycle starts after the split A ended. Likewise all the vehicles reach 
there at tl have to wait till their assigned split starts except if they expect to enter to 
the approach A. 
But when considering the vehicles reached the stop line at t2 and expecting to enter the 
approach A, the delay is little bit long compared to the rest because they have entered 
there just after the assigned split for them has ended. Hence they have to wait at the 
stop line till the next cycle starts. As same as the vehicles reach the stop line at t3 with 
the purpose to enter the approach B, have to wait till the split B starts in next cycle 
run. 
46 
MHMMMBi 
SPLIT A SPLIT B SPLIT C SPLIT D 
t 1 tg ta 
Figure 5.2- Arrivals at an Intersection over the Cycle Time. 
The following conditions were considered while calculating the delay in fixed time 
control system. 
• For the intersection shown in Figure 3.3, all four approaches have given similar 
priorities. Hence all approach green phases are set in to 30 ms each. 
• Vehicles are moving at 50 kmh average speed. 
5.3.2 Delay calculation for the approach A: 
Considering the first three sets of detector values of Table 5.1, the calculation can be 
generalized as follows [3]. 
Cy_,: Previous cycle time 
C: Current cycle time 
nA : Detector count at approach A 
SA : Split at approach A 
Since the counters are installed just after the intersection at East, and the majority of 
the vehicles coming from approach C would expected to enter to the approach A at the 
determined green phase A, each vehicle after get counted by the detector is to be on 
travel for 36.23 seconds assuming that the average speed of the vehicle is 50kmh and 
it has to travel 500 meters before reach the stop line of the approach A. 
Further, it has been assumed that the vehicles entering the approach C, which have 
been counted by the detector SCI are uniformly counted over the considered cycle 
47 
Thus, (CV_, - 36.23)^ number of vehicles have to wait at the stop line (C, , -
36.23)/2 seconds average. The vehicle counted at the beginning of the cycle C,_,, has 
to wait (C,_, - 36.23) seconds and the vehicle just counted at 36.23 seconds before 
cycle C starts will never face any delay. Hence the average delay has been considered. 
nA number of vehicles move without delay since when they reach the stop line at 
approach A, its split has started. 
{Cj_} - (C,_, -36.23) -sA}nA vehicles that is (36.23 ~sA)nA have to wait at stop 
line (C-sA) seconds till the next split starts in next cycle run. 
Table 5.4 and Table 5.5 are the tabulations calculated for the delay of intelligent 
model and the fixed time control model. Here the delay 1 and the delay 2 are meant 
for the before green delay and the after green delay. 
Cycle n, 
1.666667 
2.563692 
2.369918 
36.23) 
23.77 
26.18 
58.71 
36.23)/2 
11.885 
13.09 
29.355 
Delayl 
470.8440833 
878.5674091 
4084.392366 
sa nA 
36.15 
83.42253 
122.7854 
36.23-5 
14.54 
3.69 
(C -sA) 
40.72 
Delay2 
986.7813 
Delay total 
62.4 
1457.62542 
99.29 
878.567409 
4084.39237 
Delay % 
2335.56388 
925.392257 
Table 5.4- Delay Calculation for the Intelligent Model at Split A 2703.10547 
Delay percentages were taken for the comparison with the fixed time control model. 
For the fixed time control model, the total cycle time is never changing and it has been 
set to 120 seconds. But for the intelligent model, it is changing from time to time. 
Hence for the comparison, it has to average it to a common time horizon. Hence the 
percentage delay was taken by calculating the total delay accumulated for a 100 
seconds cycle time for both fixed time control and the intelligent model. 
48 
Cycle 
1 
2 
sa »A 
50 
76.91075 
71.09754 
N, 
1.666667 
2.563692 
2.369918 
36.23- s 
6.23 
6.23 
6.23 
36.23) 
83.77 
83.77 
83.77 
Table 5.5- Delay Ca 
(C sA) 
90 
90 
90 
36.23)12 
41.885 
41.885 
41.885 
Delay2 
934 5 
1437.462 
1328.813 
Delayl 
5847.844083 
8995.24166 
8315.346021 
Delay total 
6782.34408 
10432.7036 
9644.15896 
Delay % 
5651.9534 
8693.91967 
8036.79913 
.culation for the Fixed time Control Model at Split A 
5.3.3 Delay calculation for the approach B: 
Considering the first three sets of detector values of Table 5.1, the calculation can be 
generalized as follows. 
Cj_x: Previous cycle time 
C: Current cycle time 
nH: Detector count at approach B 
sB: Split at approach B 
SAU+\) •' Split at approach A at the next cycle CJ+] 
For the approach B, the majority from the vehicles are counted by the detector SD]. 
For this calculation, each vehicle has to move 1000 meters after got counted by the 
detector and hence has to be on travel for 72.5 seconds to reach the stop line at 
approach B. 
Considering the average delay as usual, sA nH number of vehicles will experience the 
before green delay at the stop line and have to wait (Cy_, + s /2 - 72.5) seconds. 
sR nB number of vehicles will enter the approach without delay. 
49 
Similar to the before calculation, (sc + sD)nB number of vehicles have to experience 
after green delay and have to wait ((sc+sD)/2 +sAU+i)) seconds till the next green 
phase begins. 
Table 5.6 and Table 5.7 are the tabulations calculated for the delay of intelligent 
model and the fixed time control model. 
Cycle 
sB nB 
0.763333 
2.814934 
10.64578 
0.083333 
0.256369 
0.558247 
31.62 
51.42 
80.19 
(Cj-i + SA 12-
72.5) 
-1.655 
6.18 
48.345 
(^C + ^D )/2 
+ s A(j+\) 
48.35 
77.52 
91.905 
Table 5.6-Delay Calculation for t 
21.69 
32.54 
51.81 
Delay2 
127.4023 
1021.908 
4114.206 
Delayl 
0 
51 55512258 
1398.272444 
Delay total 
127.40225 
1073.46274 
5512.47808 
Delay % 
204.137558 
1130.67489 
3648.23169 
he Intelligent Model at Split B 
When the waiting time getting negative, it meant as a zero delay since it appears when 
the previous cycle time is shorter comparing the travel time after the vehicle got 
counted. 
Cycle nB (CJ_l+SA 12-72.5) Delayl 
1 0.083333 62.5 30 156.25 
2 0.256369 62.5 30 480.6921968 3 0.558247 62.5 30 1046.713714 
SB sc + sD 
(SC+Sd)/2 
+ V -4C/+1) Delay2 Delay total Delay % 
2.5 60 60 300 456.25 380.208333 7.691075 60 60 922.929 1403.62121 1169.68435 16.74742 60 60 2009.69 3056.40404 2547.00337 
Table 5.7- Delay Calculation for the Fixed time Control Model at Split B 
50 
5.3.4 Delay calculation for the approach C: 
Considering the first three sets of detector values of Table 5.1, the calculation can be 
generalized as follows. 
Cy_,: Previous cycle time 
C: Current cycle time 
nc : Detector count at approach C 
sc: Split at approach C 
5au+d : Split at approach A at the next cycle C /+1 
sbu+\) Split a t approach B at the next cycle C /+ l 
For this calculation, the before green delay has been calculated as delay 1 and the 
delay 2 for two sets of vehicles. 
Split C is on for the approach C, in which the majority from the vehicle count from the 
intersection at West is expected to be entered. Each vehicle has to travel 300 meters 
over 21.74 seconds after getting counted by the detector . 
sA nc number of vehicles have experience a before green delay of ( Cy , + s 12 -
21.74) +s B averaged seconds at the stop line. 
sB nc number of vehicles have experience a before green delay of ( C + s /'.2 -
j-1 A B 
21.74) averaged seconds at the stop line. 
sc nc number of vehicles will never experience a delay. 
(CJ_l-sA - s B - s c ) n c number of vehicles will have after green delay and will wait 
(sD 12 + sAU+l) + sB(j+1)) averaged seconds before pass the intersection. 
51 
Table 5.10 and Table 5.11 are the tabulations calculated for the delay of intelligent 
model and the fixed time control model. 
( C M * S A 1 2 -
Cycle n c 21.74) +s B Delayl 
1 1 58.265 21.69 1263.76785 
2 1.650377 67.92 32.54 3647.524922 
3 1.706341 118.175 51.81 10447.32224 
( C m - s a + S b / 2 -
s c n c sB n c 21.74) Delay2 
21.87 9.16 21.15 193.734 
64.4472 18.12113 13.62 246.8099 
100.6912 32.53992 30.925 1006.297 
( C m - ( s D / 2 + 
s , -SR-A " 
SA(j+1) + 
s c ) n c SB(j+1) ) Delay3 Delay total Delay % 
7.28 48.395 352.3156 1809.81745 2899.884 
33.271591 77.065 0 3894.33477 4101.89 
59.636613 85.29 0 11453.6193 7580.158 
Table 5.8- Delay Calculation for the Intelligent Model at Split C 
Cycle n c 
( C h + s a 12 — 
21.74) +sB Delayl 
1 1 143.26 30 4297.8 
2 1.650377 143.26 30 7092.988303 
3 1.706341 143.26 30 7333.511692 
s c n c sB n c 
^ + S b / 2 -
21.74) Delay2 
30 30 83.26 2497.8 
49.5113 49.5113 83.26 4122.311 
51.19023 51.19023 83.26 4262.098 
SB " SC ) HC 
( S d / 2 + 
SA(j+l) + SB(j+l) ) Delay3 Delay total Delay % 
30 75 2250 9045.6 7538 
49.5112963 75 3713.347 14928.6461 12440.54 
51.1902254 75 3839.267 15434.8768 12862.4 
able 5.9- De ay Calculation for the Fixed time Control Model at Split C 
Here also zero delay will receive when negative delay time received. 
52 
5.3.5 Delay calculation for the approach D: 
Considering the first three sets of detector values of Table 5.1, the calculation can be 
generalized as follows. 
Cj_x: Previous cycle time 
C: Current cycle time 
nc : Detector count at approach C 
sc: Split at approach C 
•^O+i): sP l i1 : a t approach A at the next cycle Cj+, 
sBU+l): Split at approach B at the next cycle C ,+1 
For this calculation, the before green delay has been calculated as delay 1, delay 2 and 
the delay 3. 
Split D is on for the approach D, in which the majority from the vehicle count from 
the intersection at North is expected to be entered. Each vehicle has to travel 1000 
meters over 72.5 seconds after getting counted by the detector Sm. 
sA nD number of vehicles have experience a before green delay of (C , +s 12 -
v j ' A 
72.5) + sB +sc averaged seconds at the stop line. 
sB nD number of vehicles have experience a before green delay o f ( C , - s + s /2 -
i ' A B 
72.5)+ sc averaged seconds at the stop line. 
sc nD number of vehicles have experience a before green delay o f f C , - i + s 12 -
^ J ' A B 
72.5) averaged seconds at the stop line. 
( C j - \ ' s a ~ sB~ sc)nD number of vehicles will move without delay occurrence from 
the intersection. 
53 
Table 5.10 and Table 5.11 are the tabulations calculated for the delay of intelligent 
model and the fixed time control model. 
Cycle nD 
( C j - \ + S A 12 ~ 
72.5) + s B 
+ sc Delayl 
1 0.416667 31.03 21.69 280.433625 
2 0.544784 133.8 32.54 996.4508188 
3 0.979566 161.85 51.81 8214.100595 
s c ) sB nD 
(CH-SA + Sb/2 
-72.5)+ sc Delay2 
7.28 3.816667 -7.74 0 
-20.16 5.981734 1.91 11.42511 
-34.95 1 8.68032 39.175 731.8017 
SC nD 
-sB + sc 12-
72.5) Delay3 Delay total Delay % 
9.1125 -32.415 0 280.433625 449.3409 
21.2738343 -34.085 0 1007.87593 1061.593 
57.8041921 -18.935 0 8945.9023 5920.518 
Table 5.10-Delay Calculation for the Intelligent Model at Split D 
Cycle no 
( C j - \ + s A 12 ~ 
72.5)+ sB +5C. Delayl 
1 0.416667 60 30 750 
2 0.544784 
58.345 
30 2002.083 
3 0.979566 •58.345 30 1714.583421 
(CJ.]-sA -
sB-sc) SB nD 
+ 
Sb/2-
72.5)+ sc Delay2 
30 12.5 62.5 781.25 
30 16.34353 62.5 1021.471 
30 29.38698 62.5 836.686 
'V nD 
-
sB + sc 12-
72.5) Delay3 Delay total Delay % 
12.5 2.5 31.25 1562.5 2503.605 
16.34353 
47 2.5 40.85884 
3064.4127 
5 3227.736 
29.38698 
13 2.5 73.46745 3624.7372 2398.9 
Table 5.11-Delay Calculation for the Fixed time Control Model at Split D 
54 
With comparing the results obtained, the summary of the all tabulations are given in 
Table 5.12. 
Cycle 
time 
62.4 
94.9 
151.1 
Total Delay % 
IC 
5865.0 
7219.6 
18662.2 
Table 5.12- Improvement o 
Total Delay % 
FC 
17325.6 
25531.9 
27092.8 
btained by the Inte 
Improvement % 
66.1 
71.7 
31.1 
ligent Control Model 
It is clear that for low, medium and averaged high traffic flow data, it has been 
improved 51.6% minimization of the total vehicle seconds delay by the intelligent 
traffic control model over the fixed time control system. 
55 
Chapter 6: Conclusions and Future work 
6.1 Conclusions 
In this research, the main focus is to develop an intelligent traffic signal control 
system based on Fuzzy logics. The mechanism for the development of an intersection 
wise traffic control system has been described. 
First, one traffic four way intersection has been selected for experimenting the traffic 
load and simulation. Since the methodology described in the thesis is decentralized 
approach, one intersection results were sufficient. According to the traffic data at 
various times, the influence functions were built and the methodology was developed 
to sense the influence by the controller. The detector values are taken periodically and 
the influences are calculated accordingly. 
The Fuzzy Inference System in Matlab tool box was used to develop the fuzzy rules. 
Five fuzzy rules were developed for deciding traffic conditions and the way of 
deciding output green time was developed. 
Matlab Simulink was used to develop the simulation model of the intelligent traffic 
signal control system. Finally, the results taken from simulating the traffic load data at 
various times in both control methods, the intelligent model and the existing fixed 
time control method have been shown and the comparison of both methods with 
improvements are shown. 
At each intersection, there should be a controller. And all the rules mentioned are to 
be implemented individually, considering the traffic levels at each intersection. 
By the decentralized method developed, dependency of a central controller is avoided 
and hence it is easy for further expansions coming through construction, and the new 
intersections which would come to the grid later. Thus, adding new intersections to 
the network increases only the computation loading on neighborhood intersections 
while rest of the network remains the same. 
6.2 Future work 
The main purpose of the Intelligent Traffic Control Model is to optimize the delay 
occurring on roads and hence to minimize the heavy traffic jam at city roads. The 
method developed in this research is decentralized approach and hence to optimize the 
56 
delay at each intersection, traffic data are to be studied in each intersection in the grid. 
The influence is to be taken accordingly. 
Make of the program to implement the method described herein by a programmable 
logic controller is needed. 
Developing a mathematical model to optimize the delay and to find the equation of 
influence by mathematical modeling is suggested. Rather been depending upon the 
assumptions, influence calculation from a proper experiment through results is 
suggested. 
For finding machine learning techniques such as Neural networks, to implement 
adaptive methods to suggest alternative ways to move for the vehicles by sensing the 
huge disturbances on roads if accident or construction work occurred by the controller 
is suggested. 
57 
References 
[1] Szu-yin, Duen-Kai Chen, and Ruey-Shun Chen "Apply adaptive and cooperative 
multi agent system to urban traffic signal control", Proc. Of the 1996 IEEE 
International Conference, pp 582-589, 1996 
[2] Tavaladakis. K, Voulgaris N.C. "Development of an autonomous adaptive traffic 
control system", Paper presented at the 78th Transportation Research Board Meeting, 
Washington D.C,2000 
[3] Michael Shenoda, Randy Machemehl. "Development of a Phase by phase, arrival 
based, delay optimized Adaptive traffic Signal Control methodology with 
Metaheuristic Search", Master thesis, University of Texas at Austin, October 2006 
[4] Josefa Hernandez,Jose Cuena, Martin Molina, "real time traffic management 
through knowledge based models: The TRYS approach" , Paper presented at 12th 
European Conference on Artificial Intelligence, 2000 
[5] Pitu Mirchandani, Larry Head, "RHODES: A real time traffic signal control 
system: Architecture, Algorithms and analysis", Transportation Research Record, Part 
B, 2001 
[6] Jeanan Sha Aban, Andy Tomlinson, Benjamin Heydecker, larry Bull, "Adaptive 
Traffic control using evolutionary algorithms", Working paper from the Transport 
Operations Research Group, Newcastle University, 2000 
[7] Jim McGuire, David O' Keeffe, "The Limerick Adaptive Urban Traffic Control 
System Project", NASA Joint program on Remote Sensing, 2002. 
[8] David Olsen, "Fuzzy Logic Control and Autonomous Robotics:Investigating the 
Motorola MC68HC12 on a line following Robot", Paper presented at Wolfram 
Research Inc, 2003 
[9] Jung Hoon Lee, Beom Hee Lee," A real- time traffic control scheme of multiple 
AGV systems, a routing table approach", Proceedings of the third Annual World 
Congress on Intelligent Transport Systems USA, 2000 
[10] Tsu-Tian Lee,"Intelligent Transportation Systems", Paper presented at American 
Control Conference, 2002 
[11] Chang E.C.P, Lei J.C, Messer C. J," Arterial Signal Timming Optimization Using 
PASSER 11-87 - Micro computer User's Giude", Texas Transportation Institute, 
Texas, pp.73-85, 1998 
58 
[12] Waalace C. E„ et al," TRANSYT - 7F User's Manual", Transportation Research 
Center, University of Florida, Gainesville, Florida, pp 65-71, 1998 
[13] Chang, E. C. P., Lei, J. C„ and Messer, C. J., "Arterial Signal Timing 
Optimization Using PASSER 11-87 ," Microcomputer User's Guide, pp. 73-85,1989 
f 1 4 ] http://www.scoot-utc.com (Seimens web page) 
[15] Wallace, C. E„ et al., "TRANSYT-7F User's Manual", Transportation Research 
Center, University of Florida, Gainesville, Florida, pp.91-115, 1988. 
[16] Wann-Ming Wey,"A Study on an Urban Network Traffic Signal 
Control,"Transportation Planning Journal,Vol.29 No.4 ,pp.693~708, December 2000. 
[17] Danko A. Roozemond, Jan L.H. Rogier, "Agent control traffic lights", ESIT2000, 
pp. 14-15, September 2000, Aachen, Germany. 
[18] http://www.pwmag.com/industrv (web) 
[19] Enrique D. Ferreira, Eswaran Subrahmanian, Dietrich Manstetten "Intelligent 
agents in decentralized traffic control," IEEE Intelligent Transportation System 
Conference Proceedings,Vol 8, pp. 12-35, August, 2001 
[20] Stuart J. Russell, Peter Norvig, "Artificial intelligence/a modern approach", 
Englewood Cliffs, NJ. Prentice Hall International, pp. 81-125, 1995. 
[21] David E Lucas, Pitu B. Mirchandani, K Larry Head,"Using remote simulation to 
evaluate Real time traffic control strategies",pp.73-90, 2001. 
[22] http://www.mathwork.com (Software Manual Matlab) 
[23]Chandru Mirchandani, "Automatic Traffic control system using existing 
Infrastructure", Paper presented at the IEE International Conference on road traffic 
signaling at London, U.K. 
[24] http://www.signalsvstems.org.vt.edu (web) 
[25] http://www.opsfhwa.dot.gov/publications/adaptivecontrol (weh) 
[26] Hunt, P.B., D. I. Robertson, R. D. Bretherton, and R. I. Winton. "SCOOT - a 
traffic responsive method of coordinating signals. Laboratory Report 1014", Transport 
and Road Research Lab, Crowthorne, Berkshire, U.K. 1981 
[27] Hansen, B.G., P. T. Martin, and H. J. Perrin. "SCOOT real-time adaptive control 
in a CORSIM simulation environment", Transportation Research Record 1727. 
National Research Council, Washington, D.C, 2000. 
[28] Lowrie, P.R. " SCATS: The Sydney Co-ordinated Adaptive Traffic System", 
Paper presented at the IEE International conference on road traffic signaling at 
London, U.K, 1982 
59 
[29]Brain Dong, Graham Roff,"Adaptive Self-Confiuring Traffic control systems", 
CSE237A Final report June 11, 2004. 
[30] http://www.traffic-tech.com/pdf/ (web) 
60