Available online at www.sciencedirect.

com

ScienceDirect
Transportation Research Procedia 16 (2016) 329 – 342

2nd International Conference "Green Cities - Green Logistics for Greener Cities",
2-3 March 2016, Szczecin, Poland

The importance of automatic traffic lights time algorithms to reduce

the negative impact of transport on the urban environment
Małecki Krzysztof *
West Pomeranian University of Technology, Żołnierska Str. 52, 71-210 Szczecin, Poland

Abstract

This paper focuses on the aspect of traffic lights control in order to reduce congestion and thereby contribute to the reduction of
the negative impact of transport on the urban environment. The author presents his thoughts based on automatic traffic lights time
algorithms developed and presented in the article. The effectiveness of the algorithms was tested in a developed cellular automata
based traffic simulator.
©©2016
2015TheTheAuthors.
Authors. Published
Published by Elsevier
by Elsevier B.V. B.V.
This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the organizing committee of Green Cities 2016.
Peer-review under responsibility of the organizing committee of Green Cities 2016.
Keywords: traffic optimization; traffic lights optimization; reducing the negative impact of transport in the city; algorithms; cellular automata;
multi-agent systems;

1. Introduction

The world around is not static. It changes over time, evolves just as its individual components. In order to know
how the environment behaves to changes in any parameters, there are two paths to choose from - make changes in the
surrounding environment and observe the reaction or build a copy of the environment as close as possible to the reality
and then test its response to the introduced changes. The problem here is, however, the complexity of the environment
that surrounds us. It is not cost effective, and often physically impossible to make a copy of the environment in the
real world and then let it be destroyed. In this case, technology becomes helpful. If you cannot make a copy on a 1:1
scale, you should gather as much information about the environment as possible and recreate it virtually, i.e. build a
model of the system. Of course, it is not possible to simulate the world around us at any scale. System created in such

* Corresponding author. Tel.: +48-91-449-55-14; fax: +48-91-449-56-61.

E-mail address: [email protected]

2352-1465 © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the organizing committee of Green Cities 2016.
doi:10.1016/j.trpro.2016.11.032
330 Małecki Krzysztof / Transportation Research Procedia 16 (2016) 329 – 342

way will reproduce only its selected section. Modeling and simulation is applicable in every area of life, and recently
it has become increasingly popular in the study of traffic changes. By using more accurate mathematical models used
to describe the movement, it is possible to carry out tests and avoid errors before they are implemented by physical
interference with the existing structure of the road. Therefore, each studied solution is a new model, the behavior of
which is tested under the conditions specified for the simulation. In the case of modeling and simulation of traffic there
is, however, another variable - control of simulated traffic on modeled road section or road junction. It requires a
driver, whether in the form of a man - an expert, whose task will be to control the movement of vehicles or machine
generating signals for them. The obtained results depend to a large extent on this factor. Therefore, it is important to
choose the appropriate control algorithm and its subsequent adaptation to the prevailing conditions. This publication
shows the above issues and presents the results of research carried out using original software based on the developed
algorithms for automatic traffic lights modulation in order to increase road capacity and thus reduce the negative
effects of urban transport i.e. air and noise pollution emitted by vehicles and congestion .
Traffic simulation was made based on traffic models, supported by the theory of cellular automata, Wolfram (2002).
Traffic lights control has been enhanced by multi-agent control system by Wooldridge (2002). The entire road system
has been treated as a dynamic system by Grogono (2005), Scheinerman (2012), Wilensky and Resnick (1999), which
simply means the system evolving over time. Aspects of "green transport” have been shown in Wątróbski and Sałabun
(2016).

1.1. The negative impact of transport on the urban environment

Various publications have focuses on the estimation of the level of pollutants in transport, which include e.g.:
Gronowicz (2004), Taniguchi et al. (2001). Table 1 shows the components of exhaust gases by types of engines and
their emissions depending on the engine operation.

Table 1. Components of exhaust gases at different engine operating conditions (ZI - positive ignition engine, ZS -
compression ignition engine).
Idle speed Acceleration Constant speed Braking
Components of exhaust gases
ZI ZS ZI ZS ZI ZS ZI ZS
Carbon monoxide [%vol.] 1,3-14 0 3 0,05 3,4 0 5,5 0
Hydrocarbons [%vol.] 0,16-0,98 0,02-0,05 0,09 0,02 0,03 0,01 1,67 0,03
Nitrogen oxides [ppm] 15-45 50-68 1347 849 653 237 18 30
Aldehydes [ppm] 5-72 6-17 16 17 7 11 289 29

Source: Gronowicz (2004), p. 111.

This summary can serve as a basis for focusing attention of designers of new solutions on factors that have a greater
effect on the reduction of the amount of pollutants. The negative impact of transport and the ways of prevention and
the implementation of good practices has been discussed by Iwan and Kijewska (2014). Reduce the impact of negative
factors believed to be in electric motors, electric vehicles and hence renewable energy sources, discussed by Iwan et
al. (2014), Watrobski et al. (2016), Wątróbski and Sałabun (2016). The context of the driving techniques optimization
in order to reduce the number of acceleration and braking, which translates into a reduction of environmental load was
discussed by Forczmański and Małecki (2013). The author also knows the articles dedicated the detection of vehicles
in intelligent monitoring systems, which can translate into the assessment of traffic, vehicle counting and classification.
And this in turn allows for verification of vehicles being in separate, ecological areas of cities. These aspects are dealt
eg. by Frejlichowski et al. (2016) and Nowosielski et al. (2016).

1.2. Road junction as a dynamic system

Any road junction, and even a section of straight road can be seen as a dynamic system, because it evolves over
time (traffic varies at different times of the day, weather changes). It is possible to extract the states of this system -
e.g. values describing the number of vehicles per hour. The situation on the road (just as in the case of a system) can
 Małecki Krzysztof / Transportation Research Procedia 16 (2016) 329 – 342 331

be controlled by traffic lights or road signs that apply to moving objects (of course, except for roundabouts – Macioszek
(2013)). However, the problem may be its mathematical description. Complicated and complex models and systems
can be very difficult to develop and generate computational problems requiring large memory and high-speed
computers. One of the solutions to simplify complex calculations is to use a series of simple rules that can be calculated
quickly and executed simultaneously. Such solutions are derived from the theory of cellular automata and are called
cellular automata.

1.3. The concept of cellular automata

The simplest cellular automaton is a mathematical model consisting of three elements, presented by Kułakowski
(2000), Malarz (2008):

x a discrete n-dimensional space (grid) divided into equal cells,

x a finite k-element set of states for a single cell,
x F rule defining state of the cell at time t +1, depending on its state and state of the neighbor cells at the moment
of time t.

N-dimensional grid of identical cells evolves in discrete moments of time. Each cell in the grid can take one k state
defined by the rule:

Si (t  1) F (s j (t )), j belongs to O(i), (1)

where: O(i) – neighborhood of i-th cell, Si – state of cell, F – transition rule.

1.4. Selected traffic models based on cellular automata

The Nagel–Schreckenberg (N-Sch) model is one of the basic models of cellular automata to simulate the cars’
movement. It was developed in 1992 by Nagel and Schreckenberg (1992). This model describes the one-lane car’s
movement and is the basis for testing various traffic scenarios, Nagel et al. (1998).
Michael Schreckenberg has created a project that informs drivers from Cologne, Aachen and Bonn about traffic
jams on the road. Development and utilization of the above model enabled to predict the traffic congestion and inform
drivers about it (http://www.autobahn.nrw.de).
The most important model showing traffic within the intersection is the model by Chowdhury and Schadschneider
(1999). The authors model the traffic on one-way single-lane roads. The development of this model to the version for
two-lane and two-way roads is presented by Malecki and Iwan (2012). The new model expands the original idea of
the intersections with the mechanism of induction loops activating traffic light to eliminate congestion (keeping smooth
movement at the intersection).
Another proven model is the model developed by Biham et al. (1992). This is a simple cellular automaton model
showing the traffic into two intersecting directions. Each array cell can be occupied with vehicle traveling in one of
two directions (north or east). The vehicle moves by one cell to the chosen direction, when it is empty. In another case,
it remains in its position. This model works in the following cases:

x even courses are occupied by vehicles moving in one direction, and in odd courses vehicles moving in the second
one. This simulates simple traffic lights and illustrates the problem of simultaneous movement in both directions.
x vehicles driving in both directions are moving at the same time, if the space in front of them is empty. When two
vehicles are trying to enter the same cell, one of them is selected randomly.
x it is possible for one cell to be occupied by two vehicles, provided that they do it at the same time. Otherwise,
when a cell is already occupied, none of them can take the cell in the cellular automaton array.
332 Małecki Krzysztof / Transportation Research Procedia 16 (2016) 329 – 342

1.5. Multi-agent system and agent-oriented traffic simulation

Agent-based designing is a relatively new programming paradigm, which includes the elements of artificial
intelligence and the world of distributed systems. Applications designed as multi-agent systems consist of a collection
of objects called agents, which are characterized by e.g. autonomy, communication and proactivity. As autonomous
objects, agents are able to take the initiative to perform certain tasks without the stimulus given by the user - it can be
e.g. the internal system diagnostics and performing tasks to draw operators’ attention. Over the past few years the
agent technology has been the subject of much debate in the scientific community, but only recently it has been applied
in commercial applications - from small to complex, often critical for some projects.
Agent is a software component that has autonomy and provides interoperabilny interface for a system and/or
behaves like a human agent working for its clients, in accordance with its schedule to achieve a certain goal. It happens
that a single agent working within the system, and - if necessary - in contact with users, can act as agent system.
Typically, however, such systems consist of multiple agents. Agents in multi-agent system may have conflicting goals,
communicate with each other indirectly (by affecting the system), or directly - by negotiations. It is possible that some
agents take a joint initiative together to achieve the goal beneficial for both of them, Bellifemine et al. (2007).
One of the domains in which the application of agent technology brings benefits is traffic simulation. Growing
traffic still requires effective solutions for its control, and the widening of roads and expansion of the existing
communication network do not offer reasonable long-term solutions, not to mention the fact that they are not always
possible to implement. This forces to seek solutions to increase road capacity through intelligent traffic management,
Burmeister et al. (1997).
Traffic management is mainly a process of planning, implementing, testing and managing the flow of vehicles in
order to optimize the use of available road solutions. The effect is to meet the demand, which is associated with the
traffic flow. Therefore, more and more attention is paid to integration and combination of subsystems that support the
traffic, which are located in the infrastructure of road junction. Many components used in traffic control can be changed
to adapt them to the road conditions (traffic lights, signs, speed limits, etc.). The use of agents to make these decisions
would greatly increase the possibility to adapt the system to the road conditions. Agents can collect information from
sensors and road infrastructure on the movement of vehicles and make much quicker decisions, which could eliminate
local problems (icy road surface only on the bend forces the reduction of speed only on this bend).

1.6. Lights synchronization – “green wave”

Modern trend in the traffic management and optimization tends toward joining vehicles in streams and their control.
One way is to synchronize the traffic lights located in the analyzed road section. Assuming that the phases of traffic
lights have fixed times and distances between them are not large, it is possible to aggregate traffic by synchronizing
lights by changing the phases of green signals at the next consecutive intersections. Assuming that there is a preferred
direction with priority in terms of travel time in the particular road, it is possible to create the effect of the so-called
“green wave”. The vehicle, which starts its way from the first intersection will have green lights when heading to the
next intersection. Shifting to the next phase of the lights cycle must therefore be equal to the average travel time
between this one and the previous lights.
However, in a high traffic (or maybe because of intersecting downward streams) there is the problem of some
vehicles being before the lights and the formation of queues at the next intersection. This means that the aggregated
stream of vehicles, which in the case of an empty road would get to the traffic lights when it changes to green, will be
stopped and partly connected with a stream of vehicles, which have just moved. This interferes with the idea of a green
wave in traffic. For this reason, the idea of offsetting a cycle was developed - cycles of consecutive lights are set so as
to cause the reduction of already existing queues at further lights, and only later enabling the next aggregated stream
of vehicles to enter this section in a controlled way, Lewandowski (2007), Fisheries (2008).
The green wave is possible to be obtained not only during the synchronization of the movement of high speed in
one direction. By adjusting the speed limit and shortening or lengthening the lights cycle it is possible to obtain a green
wave in both directions. It should be noted, however, that such alignment is optimal only for the selected situation. It
may turn out that in the rush hour a cycle plan will not be adequate to the needs. It is possible, therefore, that generating
 Małecki Krzysztof / Transportation Research Procedia 16 (2016) 329 – 342 333

green wave in the direction at the entry without synchronization with the other direction will prove to be more effective
solution in the morning rush hour, Fisheries (2008).
In the case of roads forming a truss pattern, movement and its synchronization is presented in the model by Biham-
Middleton-Levine, belonging to one of the traffic models created for microscopic traffic simulation, Farhi et al. (2009).

1.7. Existing solutions

TRANSYT is a system, designed and developed initially by the Research Department of the British Government,
allowing the development of time synchronization plans. When trying to solve the problem of optimizing the operation
of a dynamic system, which is the junction, we meet with the capacity optimization using parametric diagrams - the
proportion of defining signal times through the comparison of the time required for travel and distance from the lights.
For complex connection network, however, it may turn out that the same values of travel times and distances do not
provide a sufficiently good result. Traffic on the main section can be carried out smoothly, while on the side roads
there may occur congestion caused by the large number of vehicles waiting in front of traffic lights or even not
synchronized lights, Robertson and Bretherton (1991).
TRANSYT take into account previously overlooked factor - the queue length. Its primary goal is to minimize the
amount of stops during the travel through a simulated node. In this system the measure used to determine the capacity
is cyclical flow profile - the number of vehicles passing through a road section during one whole cycle of traffic lights.
This illustrates the characteristics of the traffic for the particular light cycle. This allows to estimate the queue length
and the speed of its growth and reduction. Optimization is performed according to these criteria. The only drawback
of TRANSYT method is that it requires historical data and optimization for medium-sized road network can, according
to Robertson and Bretherton, take up the whole year.
SCOOT is an evolution of TRANSYT system described above. Instead of using historical data - it uses the on-
going data collected from the area. It adjusts the traffic control signals to the collected data. Before each phase change,
SCOOT calculates on the basis of current data, whether it is better to lengthen or shorten the current phase of up to 4
seconds. Then it calculates the cost function on the roads adjacent to this intersection, provided the cycle was changed
by 4 seconds earlier or later. Similarly cycle times for groups of intersections can be changed by a few seconds every
few minutes. Typically SCOOT 10000 performs hour optimization session for the network consisting of 1000 nodes.
Due to the fact that the optimizer uses the values adjusted during data processing, it is possible to obtain optimization
with priority paths. It is also important that in case of failure of part of the detectors, signaling plans in the area of
failure gradually return to the original program (changes are not calculated by the system in subsequent sessions). This
makes the emergency system not cause chaos in traffic. The effectiveness of this system has been confirmed by a study
carried out in Glasgow, Coventry, Worcester, Southampton and London. The current version enables to make the
priority control of public transport.
SOTL (Self Organizig Traffic Lights) uses dynamic, discrete traffic patterns, which are cellular automata, to
efficiently and using small computing resources map the dynamics of the system and its surroundings, presented in
Faieta et al. (1993). SOTL is based on a multi-agent NetLogo system, which provides mapped paths intersecting each
other along with the traffic lights located at the intersection. Initially, the vehicles are arranged randomly on the space
available and each intersecting path does not have to make the vehicle try to change the direction of travel. Vehicles
move at a maximum speed corresponding to one of its length per unit of time, but stop if they hit a red or yellow signal
and reduce their speed when there are vehicles in front of them. The study involved the comparison of the methods of
self-organization of light signals in the modeled system with static methods defining time signals is in Gershenson
(2005). SOTL methods do not require a traffic prediction in the analyzed node. They also do not also require
communication between the intersections to get coordination. They do not have fixed signal cycles.
Classical methods of traffic engineering rely heavily on linear programming. However, there are Reinforced
Learning methods to teach agents how to behave, through interaction with the environment. Each agent learns through
a consequence of its own decisions, by trial and error. Any decision of the agent is evaluated numerically. Fixed value
carries a clear result of the decision taken by the agent. Agents are focused on searching and making best decisions.
Typically RL problems are modeled as Markov Decision Problems. The agent can visit a finite number of states and
on the transition to the state it receives a numerical value indicating its reward (negative values may specify the
penalty). Each state has a variable value assigned to each other, and through appropriate actions it enables to get to the
334 Małecki Krzysztof / Transportation Research Procedia 16 (2016) 329 – 342

next state. RL methods are used to address two types of problems: prediction and control. Prediction is to know the
value of the evaluation function for the selected policy of tested solutions. On the output for each visited state, we get
the function describing what evaluation we can expect during actions beginning in that state. Control is based on the
interaction with the environment, for which the policy that maximizes the reward while passing through the state space
(best under specific terms) is looked for. At the output we get such a policy, which allows for action planning and
optimal control. This type of problem requires a solution to the problem of prediction.

2. Development of a traffic simulator for automatic traffic lights control algorithms analysis

The system, used for the purpose of this publication has been written in JAVA using JADE-supported framework
(for multi-agent systems) and XML markup language (for storing road maps of a specific structure).
The simulator enables testing of algorithms for automatic selection of light signals in the traffic lights controlling a
particular road junction. The study involved implementation of 3 methods of automatic selection of traffic light control
sequences - in such a way that the characteristics of the analyzed traffic intersection, collected after the simulation, are
compared with the model in order to obtain a more optimal value according to established criteria. It also enables
manual change of the parameters to return to the initial signal values.
At the stage of implementation of the simulation system, the following limitations were applied: the maximum size
of the analyzed maps have 74 cells (squares of an array) in wide, 33 cells in height, 10 levels above the zero level and
2 underground levels. Traffic lights are grouped by the user, who created the file with the map, the maximum
simulation duration has 100 000 cycles; the application does not include the model of overtaking vehicles.
One of the most important aspects of this tool is the handling of traffic lights at the intersection. Each traffic light
has a specific position on the map and the time values of the individual signals. It is possible to control the time of
green signal, yellow between red and green, red, yellow between red and green and cycle time offset parameter.
The program allows to collect data and display the 4 main statistics. This is done using StatisticPresenter that allow
to graphically present the selected parameters describing this dynamic system. These are:

x average speed on the road junction,

x average travel time through the intersection,
x the percentage of not moving vehicles to the moving ones,
x average travel time on the road.

2.1. Main application window

The main window of the program (Fig. 1) is opened after the application is activated. It consists of seven basic
elements:
1. The space allocated for graphic imaging simulation,
2. Simulation control panel allowing to start and pause it at any time,
3. The layer display control panel. It enables to view multilevel maps from the view of the corresponding map
layer,
4. The information panel about the cellular automaton cell,
5. Traffic light control panel. It is used to manually control the traffic light cycle selected by the user. It allows to
specify own values of cycle times and dynamically change the signal generated during the simulation. This
allows to manually block traffic or resume it in the desired direction,
6. Generator panel used to manually adjust the frequency to generate new vehicles selected by the user at the
entry, as well as the number of objects to be created by the generator. It is possible to manually generate a single
vehicle with any assigned path available for the particular generator - the path with the direction in which the
vehicle is driving,
7. Menu. It consists of three sections and is used to select options and auxiliary screens.
 Małecki Krzysztof / Transportation Research Procedia 16 (2016) 329 – 342 335

Fig. 1. The main software window of microscopic simulation and traffic lights automatic optimization [source: own research].

3. New methods for automatic control signal selection implemented in the system described

The presented system to optimize the settings for the times of traffic lights included implementation of three
different methods. These are described in the following sections of this article.

3.1. GreenLine algorithm

Evolution array of the cellular automaton is rectangular in shape with limited size, which gives accurate data on the
distance or points where the various paths intersect. Therefore, having the distance, the values of traffic light control
signals and traffic on each path, it is possible to calculate the cycle time for the analyzed situations and pass it to the
appropriate fields on the array before starting the simulation. GreenLine algorithm is the result of observation of the
movement of vehicles and simulation tests. By examining the travel time by a single path with several traffic lights,
the author made modifications to enable tuning the traffic lights in order to obtain a “green wave”. GreenLine algorithm
requires the user to provide at least one path, which is subject to change. Then, for each of the following paths the
initial cell (generator) is determined and the travel through its entire length is performed, summing the length of the
path until reaching a traffic light. Then, the average speed and average travel time to traffic light is calculated. The
same operations are performed for subsequent traffic lights. It creates a group of traffic lights, which are properly
synchronized and their signals are transmitted to vehicles approaching the individual intersections.

3.2. Genetic algorithm – rEvolution

Genetic algorithms, inspired by evolution, have the potential to solve the problems presented on a simple data
structure. In this case, they are collections of traffic control signals. Each implementation of the genetic algorithm
begins by defining and creating a population of random elements - in this case the population are different
combinations of signals in the modeled traffic lights. It is followed by evaluation of the action of each set. The sets
left in population are the ones that best cope with the problem posed. In the case of traffic simulation it is possible to
336 Małecki Krzysztof / Transportation Research Procedia 16 (2016) 329 – 342

run the simulation and collect the results to select individuals that best meet the established criteria for each set of
traffic lights. The resultant winning population is crossed between individuals, in order to replace the weakest
individuals, and then perform re-evaluation of obtained solutions. The algorithm stops when a result that meets the
assumptions is obtained or the upper limit of iteration is reached. REvolution algorithm begins by gathering values
specifying the number of evolution/iterations, the population size and the numerical value of the mutation threshold
set by the user. The gathered data are used to generate the starting population, consisting of a number of individuals
specified by the user - sets of signals across the simulated junctions for which random signal values are generated.
Each traffic light also belongs to a unique group in a way that allows to recognize the same sets of traffic lights. Then,
fast simulation used to collect statistics is performed for each individual in the population until it reaches the upper
limit of iterations specified by the user. After the tests performed, individuals are selected from all included in the
population that are the best in coping with the movement (according to the average travel time). Then, the worse
individuals are removed from the population and replaced with remuted individuals from the winners of this
evolutionary cycle.

Fig. 2. Structure of DMB CarInjection algorithm [source: own research].

 Małecki Krzysztof / Transportation Research Procedia 16 (2016) 329 – 342 337

After obtaining full population of new individuals, the algorithm performs iteration. This process is stopped after
achieving all iterations. Then, the best individual from the winners is selected and its signal values are transmitted to
the respective array cells of the cellular automaton.

3.3. Dynamic Multiagent Behaviour (DMB) algorithm

Peer to peer architecture allows to initialize communication with another agent at any time. This fits perfectly the
modeling of control signaling of dynamic system, which is the intersection. Using agents, as representatives of the
objects referring to traffic lights or vehicles, we can increase the realism of the simulation. Agent assumes the role of
traffic light controller, which communicates with the main agent determining the value of traffic control signals based
on the data collected from the traffic lights. In another case, it serves as a communication center between agents, by
transmitting information about traffic on sections controlled by the individual lights.
DMB CarInjection is the traffic lights operation algorithm in multi-agent system using momentary lack of
movement in one direction. Imagine the situation: main road with intensity k has the intersection with side road of
intensity j. If the traffic on the main road remains at the same level, the intersection uses a static plan to set the value
of traffic control signals. If, however, there is enough space on the main road to safely “insert” a few vehicles standing
in the queue on the side road, the signals at the intersection will change accordingly allowing waiting vehicles to enter
the traffic lane. The algorithm of traffic lights begins with the creation of a list of agents and their registration in the
main handling of agent-traffic controller. After registration, in each cycle of cellular automaton agents check the path
length in the front of the lights set by the user in search of vehicles. Then, a value that indicates the number of free
cells sends an agent to control the traffic, which is waiting to collect all the data and then determines whether there is
the need and if it is possible to change the signal for each group of traffic lights.
Agent responsible for the traffic light operation receives a message and, depending on its content and own state of
the signal (if the next cycle is about to come, the change will not be made); the content changes the state of the traffic
light to the time corresponding to the length of the queue of vehicles waiting to enter the lane. After the entire
procedure, the traffic light goes back to the times set in the map file.

4. Experimental study

4.1. Scenario of experimental study

Experimental study was carried out on the basis of developed road junctions presented in the form of an array
consisting of 74x33 cells of the cellular automaton. This article presents the results of the situation reflecting a set of
three intersections with roads intersecting at right angles (Fig. 3).
Each intersection has traffic lights (S) placed on the side roads. Due to the fact that the simulation applies to urban
traffic, speed limits were imposed - vehicles move up to 1,2 or 3 cells in one cycle, which means that according to the
classical NSch model, the actual speed equals to 18km/h, 36km/h or 54km/h. Vehicles are placed by the vehicle
generator (G) with a frequency set by the user and removed from the simulation by the destructors (D) at the ends of
the paths. Five different schemes of traffic were determined in order to check the effectiveness of tuning traffic lights
with the use of the developed optimization algorithms:

x low - simulation of low traffic, e.g. at night.

x normal - simulation of urban traffic, according to data taken from the region. Each direction receives a user-
defined frequency of generating vehicles.
x high - simulation of intersection with high traffic.
x vertical - the frequency of generating vehicles in map directions: north-south is much greater than east-west
direction.
x horizontal - frequency of generating vehicles in map directions: east-west/west-east is much greater than north-
south/north-south.
338 Małecki Krzysztof / Transportation Research Procedia 16 (2016) 329 – 342

In the case of testing the effectiveness of the multi-agent method, each traffic light is operated by a single agent.
Individual lights are grouped according to the direction of traffic optimization (north-south or east-west).
The measure of the effectiveness of different algorithms is the average travel time for all intersections tested. In the
case of multi-agent algorithms attention is paid to the additional parameter specifying the length of the queue of
vehicles standing in front of the traffic lights.

Fig. 3. Map showing the sequence of intersections [source: own research].

In the developed simulation software it was found that one simulation cycle lasts one second, during which the
traffic lights are handled, vehicles are moved and parameters to be present in the later statistics are collected. The
default setting of generator frequency on any map is the value of 20 vehicles per minute and 10 vehicles waiting to be
generated. These values and times of light signals can be changed by the user.

4.2. Results

This section presents the section of experimental study conducted for two-lane, two-way roads with several
intersections. At the beginning, the model was subject to parametrization by setting a time schedule of traffic lights
and determining the characteristics of the modeled road junction for different amounts of generated vehicles.

Table 2. Average travel time for manually set traffic light settings (e.g. by the expert)
Simulation
Scheme
1 2 3 4 5 6 7 8 9 10 avg.
Low 24.64 24.53 24.32 24.66 24.97 25.34 24.67 34.39 25.74 25.45 25.87
Levels 31.54 31.96 31.90 32.35 31.67 30.36 31.40 32.17 32.65 31.33 31.73
High 37.75 37.60 38.34 38.36 38.96 37.67 39.17 37.10 38.87 37.22 38.10
Normal 32.79 32.05 32.98 33.37 32.56 34.85 33.46 31.85 34.26 33.64 33.18
[source: own research]
 Małecki Krzysztof / Transportation Research Procedia 16 (2016) 329 – 342 339

The parameters passed to the simulation were consistent with the schemes and study scenario. 60 vehicles to be let
through the node were determined for each of the possible directions. With the results of the initial model (Table 2)
set, the study began with automatic selection of the sequence of traffic light signals in order to obtain similar or even
better results.
GreenLine algorithm. Four series of ten simulations each were performed for specific frequency schemes. The
results are presented in Table 3.

Table 3. Simulation results for each of the frequency schemes of generating vehicles using GreenLine method
Scheme Simulation
1 2 3 4 5 6 7 8 9 10 avg.
Low 25.12 25.34 26.66 25.36 25.49 25.76 25.34 25.93 26.14 26.06 25.72
Levels 37.67 37.66 37.35 36.32 36.25 36.19 37.74 37.62 36.53 37.43 37.11
High 44.27 45.35 45.17 44.58 45.92 44.36 44.47 44.19 45.64 44.33 44.83
Normal 36.39 36.57 36.37 37.65 36.29 36.19 35.84 35.52 36.65 36.73 36.42

[source: own research]

GreenLine method well chose the time values of lights in case of low frequency of vehicle appearance. With
increased congestion, vehicles did not keep up with leaving the queue in one cycle of traffic lights. The simulation
showed, however, smooth movement in the optimized direction (west-east), allowing the whole columns of cars to
travel without stopping.

Table 4. Simulation results for each of the frequency schemes of generating vehicles
using the rEvolution method
Parameters Scheme
i m p Low Normal High Vertical Horizontal
10 10 10 38.24 44.19 58.84 43.39 53.35
10 50 15 36.59 43.59 52.41 41.12 52.19
10 100 30 32.40 47.30 64.88 45.15 49.46
100 10 10 28.92 39.27 53.35 39.35 63.77
100 50 10 37.74 41.18 53.77 42.79 51.33
100 100 20 29.93 43.30 56.47 41.13 62.24
1000 100 30 30.02 42.28 51.35 42.45 54.33
[source: own research]

Scenario of rEvoultion method assumes a random set of control signals at each of the intersections. This method
was calibrated from the beginning, generating not always synchronized sequences, as in the case of a single junction.
This article presents only certain combinations of population size and iteration value, as well as mutation rates. The
study involved the determination of coefficient i setting the number of iterations to 100 and 1000 iterations, ratio m
defining the chance of mutation = 10, 50 and 100 percent, and ratio p specifying the size of the population = 10, 20
and 30 individuals. Combinations were selected from these factors representing, according to the author, the greatest
diversity and allowing to examine the effectiveness of the control based on genetic algorithm. Table 4 presents selected
combinations with average times of passing through a node in each of the five frequency schemes of generating
vehicle.
Sequences generated using this algorithm often caused congestion letting vehicles located on any road to enter the
intersection (without direction synchronization).
Scenario for DMB algorithm testing, using agent technology, requires changes in the frequency of generating
vehicles for individual schemes. The new values are presented in Table 5. The results of this algorithm are presented
in Table 6.
340 Małecki Krzysztof / Transportation Research Procedia 16 (2016) 329 – 342

Table 5. The frequency values of generators for each

of the schemes
Direction
Scheme
N-S W-E
Low 5 5
Vertical 20 5
Horizontal 5 20
High 20 20
Normal 10 10
[source: own research]

Table 6. The average travel time for DMB CarInjection algorithm

Simulation
Scheme
1 2 3 4 5 6 7 8 9 10 avg.
Low 14.20 14.65 14.66 14.34 13.65 14.21 15.18 14.34 14.11 15.38 14.47
Vertical 15.02 15.58 15.35 15.29 15.11 15.01 15.94 15.69 15.85 15.46 15.43
Horizontal 15.31 15.46 15.85 15.49 15.23 15.85 15.27 15.68 15.55 15.76 15.54
High 31.92 31.94 32.96 31.36 33.35 33.75 32.55 32.36 31.55 32.66 32.44
Normal 14.62 15.83 14.24 15.33 15.87 14.51 14.35 14.36 14.27 14.75 14.81

[source: own research]

The results of this algorithm, with the settings causing the large volume of vehicles generated on each of the paths,
are similar to the initial (determined by an expert) traffic light plan. If the traffic on one of the roads is not continuous,
the algorithm lets vehicles waiting to enter even though in the normal traffic light cycle they would have to wait a
predetermined time interval. This is reflected by the average travel times in the low and normal frequency scheme of
generating vehicles. High frequency scheme (high traffic) will block the ability to “insert” vehicle into the traffic
intersecting its direction – traffic lights follows its signaling scheme, so the time of passing through a node increases
significantly.

Table 7. Summary of results for different methods of determining the time of traffic light and different
frequencies to generate new vehicles.
Vehicle generator operation scheme
Method
Low Normal High Vertical Horizontal
Values set by an expert 25.87 33.18 38.10 N/A 31.73
GreenLine 25.72 36.42 44.83 N/A 37.11
rEvolution 28.92 39.27 51.35 39.35 49.46
DMB CarInjection 14.47 14.81 32.44 31.93 15.54

[source: own research]

Table 7 shows the results of each of the modeled traffic solutions, using the manual setting of traffic lights, and the
simulation results with the values determined by the automatic method for selecting the time of the individual lights.
Developed results classify developed methods as effective, because they generated results similar to those obtained
by manual traffic light adjustments. In addition, the algorithms tested under a large load did not cause congestion and
blocking of the tested intersection. For less traffic it is worth to pay attention to multi-agent method (DMB
CarInjection), which allows for removing the queue of vehicles on the side road so efficiently that the average travel
time, measured during the entire simulation, was almost 50% less than the initial time on manual set of traffic lights.
 Małecki Krzysztof / Transportation Research Procedia 16 (2016) 329 – 342 341

This applied to the traffic level in low and high schemes and forcing one of the directions of increased traffic. This
method ceased to be effective in case of a very large number of vehicles on each of the directions. It should be noted,
however, that the current traffic regulations do not allow to interrupt the traffic light cycle, making it impossible, at
least for the moment, to test such solutions in real road conditions.
GreenLine method generates results most similar to the results obtained by manual adjustment of traffic lights. The
most unpredictable results were obtained using a method based on genetic algorithm. This method required the longest
time for setting values and further results often resulted in congestion in the tested road junction.

5. Conclusions

The study involved the development of a simulator to test the timing of traffic light. It included the implementation
of basic road traffic models (N-Sch and Ch-Sch), enriched with parameter ‘nervousness of the vehicle driver’. This
parameter causes the vehicle to look for other paths in the event of prolonged stop. Mechanisms to handle the traffic
lights were prepared allowing for free setting of traffic control signals. Statistics module were developed and
implemented, which allows for the presentation of simulation results, e.g. the average travel time of the selected path
and the average travel time through the road junction. Module of automatic selection of traffic light control sequences
was developed with three methods:

x GreenLine - consisting of a static time calculation based on values read from the simulated area.
x rEvolution - involving the interpretation of a set of traffic lights as chromosomes and performing a genetic
algorithm,
x DMB CarInjection involving the use of agent technology, and recognizing the possibility of entering vehicles
waiting at the intersection into the direction intersecting the path (this aspect is very future-oriented and suitable
for use in the event of future autonomous vehicle, which would react to the signal immediately, that is much faster
than the average driver would react).

The developed system is an open system, which allows for the future development for more accurate movement
simulation. The proposed development directions include:

x using existing digital maps reflecting the actual road situation,

x adding the overtaking rules for more detailed mapping of reality,
x distinguishing vehicles into trucks, buses and special vehicles (ambulance, police, fire brigade).

Acknowledgements

The author wishes to thank Mr. Maciej Szmajdziński, without whom this system would be probably not developed.
This paper was financed under the project "Analysis of information needs of heterogeneous environment in
sustainable urban freight” by the Polish National Science Centre, decision number DEC-2012/05/B/HS4/03818.

References

Wolfram, S., 2002. A new kind of science, Wolfram Media.

Wooldridge, M., 2002. An Introduction to Multiagent Systems, Wiley.
Grogono, P., 2005. Dynamic systems. Technical report, Concordia University.
Scheinerman, E.R., 2012. Invitation to dynamical systems, Dover Publications.
Wilensky, U., Resnick, M., 1999. Thinking in Levels: A Dynamic Systems Approach to Making Sense of the World, Journal of Science Eductaion
and Technology 8, No.1.
Wątróbski, J., Sałabun, W., 2016. Green Supplier Selection Framework Based on Multi-Criteria Decision-Analysis Approach. In: Setchi, R.,
Howlett, R. J., Liu, Y., & Theobald, P. (eds.) Sustainable Design and Manufacturing 2016. Springer International Publishing, 361-371.
Gronowicz, J., 2004. Ochrona środowiska w transporcie lądowym. Wyd. II, WITE, Poznań-Radom.
Taniguchi, E., Thompson, R. G., Yamada, T., van Duin, R., 2001. City Logistics. Network Modelling and Intelligent Transport Systems, Pergamon,
Oxford.
342 Małecki Krzysztof / Transportation Research Procedia 16 (2016) 329 – 342

Iwan, S., Kijewska, K., 2014. The Integrated Approach to Adaptation of Good Practices in Urban Logistics Based on the Szczecin Example.
Procedia Social and Behavioral Sciences 125, 212-225.
Iwan, S., Kijewska, K., Kijewski, D., 2014. Possibilities of Applying Electrically Powered Vehicles in Urban Freight Transport. Procedia Social
and Behavioral Sciences 151, 87-101.
Watrobski, J., Ziemba, P., Wolski, W., 2015. Methodological aspects of Decision Support System for the location of renewable energy sources. In:
Computer Science and Information Systems (FedCSIS), Federated Conference on. IEEE, 1451-1459.
Watróbski, J., Sałabun, W., 2016. The Characteristic Objects Method: A New Intelligent Decision Support Tool for Sustainable Manufacturing. In:
Setchi, R., Howlett, R. J., Liu, Y., & Theobald, P. (eds.) Sustainable Design and Manufacturing, Springer International Publishing, 349-359.
Forczmański, P., Małecki, K., 2013. Selected Aspects of Traffic Signs Recognition: Visual versus RFID Approach, Activities of Transport
Telematics, Communications in Computer and Information Science 395, 268-274.
Frejlichowski D., Gościewska K., Forczmański, P., Nowosielski, A., Hofman, R., 2016. Applying Image Features and AdaBoost Classification for
Vehicle Detection in the ‘SM4Public’ System, Proceedings International Conference Image Processing and Communications IP&C 2015, Image
Processing and Communications Challenges 7, Advances in Intelligent Systems and Computing 389, 81-88.
Nowosielski, A., Frejlichowski, D., Forczmanski, P., 2016. Automatic Analysis of Vehicle Trajectory Applied to Visual Surveillance, 8th
International Conference on Image Processing and Communications (IP and C), Advances in Intelligent Systems and Computing 389, 89-96.
Macioszek, E., 2013. Analysis of the Effect of Congestion in the Lanes at the Inlet to the Two-Lane Roundabout on Traffic Capacity of the Inlet.
[w:] J. Mikulski (ed.) Activities of Transport Telematics. Communications in Computer and Information Science 395. Springer-Verlag Berlin
Heidelberg, 97-104.
Kułakowski, K., 2000. Automaty komórkowe, Akademia-Górniczo-Hutnicza.
Malarz, K., 2008. Automaty komórkowe, Akademia-Górniczo-Hutnicza.
Nagel, K., Schreckenberg, M., 1992. A cellular automata model for freeway traffic. Journal de Physique I 2, 2221-2229.
Nagel, K., Wolf, D. E., Wagner P., Simon P. M., 1998. Two-lane traffic rules for cellular automata: A systematic approach, Physical Review E
58(2), 1425–1437.
Chowdhury, D., Schadschneider, A., 1999. Self-organization of traffic jams in cities: effects of stochastic dynamics and signal periods. Physical
Review E 59, 1311-1314.
Malecki, K., Iwan, S., 2012. Development of Cellular Automata for Simulation of the Crossroads Model with a Traffic Detection System.
Telematics in the Transport Environment 329, 276-283.
Bellifemine, F.B., Caire, G., Greenwood D., 2007. Developing Multi-Agent Systems with JADE, Wiley.
Burmeister, B., Haddadi, A., Matylis, G., 1997. Application of Multi-Agent Systems in Traffic and Transportation, IEE Proceedings on Software
Engineering 144, 51 60. ͒
Lewandowski, K., 2007. Idea Zielonej Fali dla pojazdów drogowych we Wrocławiu , Politechnika Wrocławska. ͒
Rybacki, B., 2008. Modelowanie i optymalizacja ruchu miejskiego przy użyciu wybranych technik , Akademia-Górniczo-Hutnicza. ͒
Farhi, N., Goursat, M., Quadrat J., 2009. About Dynamical Systems Appearing in the Microscopic Traffic Modeling, arXiv 0911.4672. ͒
Biham, O. Middleton, A.A., Levine D., 1992. D.: Phys. Rev. A 4 6 6124.
Robertson, D., Bretherton, R., 1991. Optimizing networks of traffic signals in real time - the SCOOT method, Vehicular Technology 40, No.1. ͒
Faieta, B., Huberman, B.A., 1993. Firefly: A Synchronization Strategy for Urban Traffic Control, Technical Report SSL-42, Xerox PARC. ͒
Gershenson, C., 2005. Self-organizing Traffic Lights, Complex Systems Journal 16, 29-53. ͒