(IJACSA) International Journal of Advanced Computer Science and Applications,

Vol. 11, No. 3, 2020

Optimizing Genetic Algorithm Performance for

Effective Traffic Lights Control using Balancing
Technique (GABT)
Mahmoud Zaki Iskandarani
Faculty of Engineering
Al-Ahliyya Amman University, Amman, Jordan

Abstract—Genetic Algorithm (GA) is implemented and light signal, would not greatly improve traffic flow as
simulation tested for the purpose of adaptable traffic lights important parameters, such as, que length and vehicle speed are
management at four roads-intersection. The employed GA uses not taken into account in a standard control algorithm [5].
hybrid Boltzmann Selection (BS) and Roulette Wheel Selection
techniques (BS-RWS). Selection Pressure (SP) and Population Traffic congestion is a critical issue in affecting the lives of
(Pop) parameters are used to tune and balance the designed GA the society. Many areas suffered from long term
to obtain optimized and correct control of passing vehicles. A socioeconomic damage owing to growing traffic congestion.
very successful implementation of such parameters resulted in To resolve urban congestion, conventional alternatives of
obtaining minimum number of Iterations (IRN) for a wide increasing road capacity through road network expansion is
spectrum of SP and Pop. The algorithm is mathematically limited in effectively reducing congestion, as large capital
modeled and analyzed and a proof is obtained regarding the investment and stakeholders support is needed.
condition for balanced GA. Such Balanced GA is most useful in
traffic management for an optimized Intelligent Transportation Intelligent Transport System (ITS) is used for the last few
Systems, as it requires minimum iterations for convergence with years in an effort to enhance efficiency and effectiveness of the
faster dynamic controlling time. existing transportation infrastructure by employing various
sensor and communication technologies. Adaptive traffic
Keywords—Genetic algorithm; traffic lights; intelligent signal control systems try to use the principles provided by
transportation systems; correlation; roulette wheel selection; artificial intelligence in an attempt to reduce congestion and
boltzmann selection; selection pressure; population provide safer traffic operations for both vehicles and
pedestrians.
I. INTRODUCTION
As a result of population growth, increased level of II. BACKGROUND
pollution, people migration to urban areas from rural and to Excessive Traffic density on the roads is a critical issue, as
cities from urban areas, traffic congestion issue became more it leads to congestion. This is caused by rise in the number of
critical and pressing. Hence, Traffic lights optimization using vehicles and due to expansion and urbanization. Limitations on
adaptive and intelligent algorithms in conjunction with smart development and building of new highways and roads, initiated
sensing is a must under these circumstances. This comes under the need to optimize the use of existing infrastructure to
design optimization, where smart algorithms such as genetic achieve optimal flow of traffic. In addition, important time
and fuzzy logic algorithms are employed to improve an wasted because of traffic congestion, will implicitly affect
existing transportation network as a function of increasing level productivity and performance, and thus affects people’s lives,
of traffic flow levels, resulting in traffic congestion, delays, both economically and socially.
higher fuel consumption, air pollution and increase probability
of traffic incidents [1-2]. Traffic light signal management and control has a marked
impact on the efficiency and effectiveness of urban
In conjunction with population growth, a marked increase transportation systems. Conventional Traffic light signals are
in the automotive industry is witnessed, where millions of pre-programmed, pre-timed signals. Pre-timed control
vehicles are put on an existing roads. Monitoring the vehicular comprises a series of fixed duration intervals that are repeated
activities is an important issue to the transport authorities. Such continuously. Advanced Traffic light signals can operate in two
large number of vehicles have a major impact on the ways:
environment and daily living of people using these roads due to
congestion and delays [3-4]. 1) Actuated mode: Actuated Traffic Light Signals detect
and respond to the presence of vehicles or pedestrians at the
The congestion issue forms a bigger problem on lanes and intersection. They are supported by detectors within the
roads that are not structured appropriately, where it has a major
intersection and the necessary control functionality to respond
effect on vehicular movements in a city. In particular at the
intersections or with every region of traffic signal, especially to traffic density and demand, in order to affect the signal
during the peak hours. Shortening the duration of the traffic cycles times dynamically.

283 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 11, No. 3, 2020

2) Adaptive mode: Adaptive Traffic Light Signals control control mechanism, which is adaptable in nature and intelligent
system continuously calculates optimal signal timings based in behavior.
on detected volumes and dynamically implement them. The Selection process is a critical part in genetic algorithms,
system smartly and efficiently responds to the abrupt and fast whereby chromosome is chosen from the available
changes in dynamic traffic conditions. It uses data from generation’s population to be included in the next generation.
vehicle detectors so that traffic signal optimization is Fitness function based process selects the best chromosomes,
achieved. which is used to improve chances of individual survival.
Transportation network optimization (TNO) applied to The employed probability of selection process is based on
traffic signal control, is implemented while considering route two known algorithms:
choice pattern of network users. It also include minimization of
1) Boltzmann Selection (BS): The algorithm is based on
travel time and avoidance of new road construction.
Conventional traffic lights operates using a constant switching simulated annealing. Annealing is based on cooling such that a
cycle regardless of traffic load. Such static mechanism will not low energy state is reached. During the process, heating until
allow for variation of traffic load, events, emergency melting is reached at a high temperature, through which,
conditions, or general road incidents. Thus, there is a need for random movement is realized. The temperature is slowly
smart and adaptive algorithm that dynamically control traffic cooled until the minimum energy states achieved. The
signals not only locally, but also all over the network for both equivalence between the optimization process in this work and
vehicles and pedestrians. Such a critical solutions will carry out the Boltzmann Simulated Annealing (SA) is the need to
functional synchronization between traffic signals, in order to achieve stable levels with better new solutions (optimized
achieve a measurable reduction in congestion and delay levels, traffic signal timing). Boltzmann Selection (BS), general
in addition to less pollution and safer driver and pedestrian
expression is presented in equation (1).
roads. Such an objective can be achieved using genetic
algorithm (GA) [6-10].  yi 1  yi 
PSelection  i 1    exp 
 
...(1)
Genetic algorithms are used to imitate the processes of  Ti 1  (1)
natural selection, where the best individuals have more
probability to survive and their genes will be part of the Where;
creation of one or more offspring. Such process is repeated
with the output of each new offspring is more fitted to survive yi: Fitness function value for initial solution
than its parents. In the last few years, genetic algorithms found
to present a suitable approach to complex transportation
problems, as they are considered search algorithms that operate yi+1: Fitness function value for the new solution
on the principles of natural selection. They determine a number
of potential solutions within a population, with an encoding T: New artificial temperature
process that result in an optimized solution. In transportation,
genetic algorithms are used to optimize and adapt the green β: Normalization factor
interval response as a function of traffic density based on
vehicles count [11-15]. In the algorithm used in this work, the rate of selection is
managed by a continuously changing Delay Time (DT)
Selecting the best parameters of a genetic algorithm, so as parameter, which is equivalent to temperature in Boltzmann
to obtain good results to optimize its performance, is very algorithm (simulated annealing). Initially Delay Time (DT) is
important to its effectiveness. Crossover, mutation rate and high. So, one DT (Shortest Delay Time (SDT)) parameter will
population size are the most influencing control parameters as be at first high and decreases, with another DT (Longest Delay
reported by previous works [16-20]. However, pressure Time (LDT)) providing Delay Time as a function of increase in
selection and population size in correlation is a new approach the testing population. (SDT) decreases gradually which
in balancing and GA algorithm optimization. increases the effect of Selection Pressure (SP). This results in
determining and mapping the search space. In Boltzmann
In this paper an investigation is carried out regarding
selection, the probability of selecting best value is high with
selection pressure and population parameters within a GA
lower execution time.
algorithm used for intelligent traffic lights control and
management, in order to obtain best adaptable performance 2) Roulette Wheel Selection (RWS): In this algorithm,
under dynamically changing traffic density. The two selection of the fittest is carried out. The initial part of the
parameters and their correlative effect on the performance of selection process based on stochastic selection from one
the employed GA algorithm is mathematically analyzed, population to create the basis for the next population. In this
simulated and results discussed.
process, the fittest have a better chance (probability) of
III. METHODOLOGY survival than weakest ones. Thus, the fittest will move
forward to the mating region to prepare for the next
The main objective of this work is to optimize the designed
genetic algorithm in order to produce an optimum traffic lights population. The process is shown in Fig. 1.

284 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 11, No. 3, 2020

Boltzmann Selection (BS) and varies with Population

exploration. Equation (3) can be rearranged as in equation (4)
  SDT  
S Pr ob  exp   SP   ...(4)
Roulette   LDT   (4)
Wheel

SDT and LDT are important parameters in the genetic

algorithm computations, as they are related to the used fitness
function. They are vital for the designed algorithm in
determining the value of SProb, and thus, affect its controlling
Fig. 1. Roulette Wheel Selection.
mechanism and number of Passing Vehicles (PV). Hence,
The number of times the roulette wheel runs is SProb is a function of both the Selection Pressure (SP) and the
proportional to the size of the desired population. Whenever dynamic ratio of the genetic algorithm times that depends on
the wheel stops fittest individual will have a good chance of the population value, which is related to other parameters such
as offsprings, mutants and crossover levels.
being selected for the next population and subsequent mating
region. Thus, selection of SP and Pop, will affect SProb, which in
turn affects the algorithm computations and Vehicles Passing
The algorithm applies the expression in equation (2):
Rate (VPR). So, balancing and optimization of the two values
  will result in a tuned, balanced and efficient GA for traffic
  signals and traffic control and management.
PSelection  m    n m ...(2)
y
 
  yk
Two conditions associated with Selection Pressure:

 k 1  (2) 1) High Selection Pressure Value: Early convergence.
2) Low Selection Pressure Value: Late convergence.
Where;
Thus, an optimum selection pressure value is needed for
optimal control of traffic lights, which is a dynamic,
n: Population Size changeable value function of traffic density, which is computed
in the designed algorithm as in equations (5) and (6) for each
m: Number of times the wheel rotated road i.
The combined Boltzmann Selection-Roulette-Wheel TPi  RCi  VPi ...(5)
Selection (BS-RWS), for genetic algorithms (GAs) is based on (5)
both entropy and importance sampling methods. It naturally
 RCi  PVi  
leads to adaptive fitness in which the fitness function does not TPRi   ...(6)
stay fixed but dynamically varies.  RCi  (6)
Two important controlling parameters are used in
optimizing and validating the used genetic algorithm for a four- Where;
traffic lights intersection covering four roads (Rd.1, Rd.2, Rd.3,
Rd.4), each having a capacity of 60 vehicles: TP: Traffic Pressure
1) Selection Pressure (SP)
2) Population (Pop) TPR: Traffic Pressure Ratio
Selection Probability (SProb) for traffic lights control,
PV: Passing Vehicles
which in effect manages the number of vehicles passing from
road to road at the four-road junction is governed by equation
(3), which is based on Boltzmann general expression in RC: Road Capacity
equation (1).
i: Road at the 4-road intersection (i=1, 2, 3, 4)
  SP x SDT 
S Pr ob  exp  ...(3) PV is related to both SP and Pop, subsequently is affected
 LDT  (3) by SProb, which is a function of SDT and LDT and SP. Equal
PV for same TP per road should be equal for the GA algorithm
Where; to be considered balanced and optimized. The general process
of optimization is shown in Fig. 2.
(SDT): Shortest Delay Time (Best Time), which is a
dynamic parameter that varies with number of iterations (IRN). The idea is to optimize the GA performance using both SP
and Pop to match the balancing data shown in Table I.
(LDT): Longest Delay Time (Worst Time), which is a
dynamic parameter that corresponds to Temperature in

285 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 11, No. 3, 2020

1000
900
800
700

Iterations (N)
600
500
400
300
200
100
0
0 2 4 6 8 10 12
Selection Pressure (SP)

Population=200 Population=400 Population=600

Population=800 Population=1000

Fig. 3. Iterations (IRN) as a Function of SP Per Specific Pop.

1000
900
800
700

Iterations (N)
600
500
400
300
200
100
0
0 200 400 600 800 1000 1200
Population (P)

SP=2 SP=4 SP=6 SP=8 SP=10

Fig. 4. Iterations (IRN) as a Function of Pop Per Specific SP.

Fig. 2. GA Optimization Algorithm.

TABLE. I. GA BALANCING REFERENCE VALUES

V. ANALYSIS AND DISCUSSION
The analysis process is based on the stability criteria,
Vehicles Passing Times (sec) No. of Passing Vehicles
POPi whereby, a correlation between three parameters (SP, Pop,
Rd.1 Rd.2 Rd.3 Rd.4 Rd.1 Rd.2 Rd.3 Rd.4
Iterations) is carried out, with final correlation to smooth and
SP correct traffic lights control at the junction, such that the four
2 17 57 57 17 7 23 23 7 roads will pass the correct number of vehicles. This is a
4 17 57 57 17 7 23 23 7 calibration and load balancing measure, which is based on
6 17 57 57 17 7 23 23 7 initial conditions of equal traffic pressure for opposing roads.
8 17 57 57 17 7 23 23 7
Table II shows data represent the Iteration parameter as a
10 17 57 57 17 7 23 23 7
function of the relationship between SP and Pop, whereby
Iterations are dependent on the two parameters (SP, Pop) with
IV. RESULTS
each SP level is fixed for a variable levels of Pop. The general
Table II shows the number of Iterations (IRN) required for form of dependency is described by equation (7).
each Selection Pressure Value to achieve balanced and
optimized GA performance as a function of the relationship IRN  f SP, Pop...(7) (7)
between Population (Pop) and Selection Pressure (SP), with
plots representing the data shown in Fig. 3 as a function of SP Now, Pop affects LDT, hence affecting SProb. Also, SP
and in Fig. 4 as a function of Pop. affects SProb according to equation (4). Thus, IRN, will affect
SProb and by using both equations (4) and (7), equation (8) is
TABLE. II. IRN AS A FUNCTION OF BOTH SP AND POP obtained.
Iterations (IRN) Population (Pop) S Pr ob  f IRN ...(8) (8)
SP 200 400 600 800 1000
The PV parameter depends on SProb, hence, can be
2 650 350 300 50 50
described by equation (9).
4 500 200 50 150 50
6 300 50 200 50 50
PV  g S Pr ob  g  f IRN ...(9) (9)
8 450 50 200 50 50
10 950 400 400 400 50

286 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 11, No. 3, 2020

Then, for each opposite and equal TP roads, the difference

in the rate of change should be zero for balanced GA
algorithm. Thus; for i (1 to 4):
 PVi   gi  f IRNi   0...(10)
(10)
The condition in equation (10) will only apply if the
number of iterations in the GA algorithm is the same for each
SP and Pop to enable same number of vehicles to pass on each
road as a function of TP. This condition is valid for stable GA
algorithm. Hence, when achieved, the GA algorithm is
balanced and optimized as equation (11) shows.
Fig. 6. {2,1000} LDT Curve.
 IRNi  0 ...(11)
(11)
From Table II, it is clear that the Steady State (SS) values is
reached for the GA algorithm controlling four traffic lights
with one intersection at Pop=1000, with minimum number of
Iterations of 50, which is also constant for all levels of SP.
Such convergence, appears in Fig. 2.
Table III, presents data that shows effect of Pop on IRN in
order to reach SS, with Fig. 3 showing the convergence to an
SS described by the pairing in equation (12):

SS  ( Pop, IRN ) ...(12) (12)

From Table II, it is realized that for some SP, Pop values Fig. 7. {10,1000} LDT Curve.
oscillation occurs during tuning and balancing of the GA
algorithm. The two non-oscillatory values are shown in
Table III, with SP=2 having early convergence to minimum
number of iterations. Non-oscilatory SP, Pop pairs are prefered
in the balancing process to other values, eventhough the
convergence condition achived with intermediate GA
oscillations.
Fig. 5 to 8 show SDT and LDT for the two non-oscillatory
SP values.

TABLE. III. NON-OSCILLATORY SP AND POP

Iterations (IRN) Population (Pop)

Fig. 8. {10,1000} LDT Curve.
SP 200 400 600 800 1000
2 650 350 300 50 50 VI. CONCLUSIONS
10 950 400 400 400 50 The obtained condition for number of interactions (IRN) in
this work and subsequent steady state condition specified, is an
important achievement, as it uncovers conditions of constant
and low number of iterations necessary to obtain balanced, and
optimized GA control algorithm used in the management of
traffic lights signaling process. The used technique locally
weights the processes using both Pop and SP correlation,
which enables optimized functionality and better forecasts to
achieve optimal dynamic traffic modeling. The balanced GA
algorithm proposed in this work offers tangible advantages.
Future work requires the extension of such application of
balanced GA algorithm is recommended to cover multi-section
traffic control, whereby each four-road, single intersection is
regarded as a single parameter and correlated with other similar
structure to cover more complex arrangement.
Fig. 5. {2,1000} SDT Curve.

287 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 11, No. 3, 2020

REFERENCES [11] A. Potnurwar, S. Aote, V. Bongirwar, “Design of Traffic Volume

[1] J. Chen, Y. Yu, Q. Guo , “Freeway Traffic Congestion Reduction and Forecasting based on Genetic Algorithm,” International Journal of
Recent Technology and Engineering, vol. 8, no. 2, pp. 4264–4268, 2019.
Environment Regulation via Model Predictive Control,” Algorithms,
vol. 12, no. 220, pp. 1–23, 2019. [12] A. EL Idrissi, C. Tajani, M. Sabbane, “New Crossover Operator for
Genetic Algorithm to Resolve The Fixed Charge Transportation
[2] G. Jia , R. Ma, Z. Hu , “Review of Urban Transportation Network
Problem,” Journal of Theoretical and Applied Information Technology,
Design Problems Based on CiteSpace,” Journal of Theoretical and vol. 98, no. 8, pp. 1607–1617, 2017.
Applied Information Technology, vol. 2019, ID. 5735702, pp. 1–22,
2019. [13] W. Wen-jing, “Improved Adaptive Genetic Algorithm for Course
Scheduling in Colleges and Universities,” iJET, vol. 13, pp. 29–42,
[3] A. Goyal, M. Singh, A. Aeron, “Simulation of Traffic Optimization to 2018.
Reduce Congestion,” International Journal of Innovative Technology
and Exploring Engineering, vol. 8, no. 11, pp. 3780–3783, 2019. [14] Y. Wang , X. Yang , H. Liang , Y. Liu, “A Review of the Self-Adaptive
Traffic Signal Control System Based on Future Traffic Environment,”
[4] S. Yang, Y. Ji, D. Zhang, J. Fu, “Equilibrium between Road Traffic Journal of Advanced Transportation, vol. 2018, ID. 1096123, pp. 1–12,
Congestion and Low-Carbon Economy: A Case Study from Beijing, 2018.
China,” sustainability, vol. 2019, no. 11, pp. 1–22, 2019.
[15] Q. Tang, B. Friedrich, “Design of Signal Timing Plan for Urban
[5] B. Sony, A. Chakravarti, M. Reddy, “Traffic Congestion Detection Signalized Networks including Left Turn Prohibition,” Journal of
using Whale Optimization Algorithm and Multi Support Vector Advanced Transportation, vol. 2018, ID.1645475, pp. 1–16, 2018.
Machine,” Journal of Theoretical and Applied Information Technology,
vol. 7, no. 6C2, pp. 589–593, 2019. [16] S. Srivastava, S. Sahana, “Application of Bat Algorithm for Transport
Network Design Problem,” Applied Computational Intelligence and Soft
[6] Q. Bing , D. Qu , X. Chen , F. Pan, J. Wei, “Short-Term Traffic Flow Computing, vol. 2019, ID. 9864090, pp. 1–12, 2019.
Forecasting Method Based on LSSVM Model Optimized by GA-PSO
Hybrid Algorithm,” Discrete Dynamics in Nature and Society, vol. [17] Z. Cakici, Y. Sazi Murat, “A Differential Evolution Algorithm-Based
2018, ID. 3093596, pp. 1–10, 2018. Traffic Control Model for Signalized IntersectionsAdvances in Civil
Engineering, vol. 2019, ID. 7360939, pp. 1–16, 2019.
[7] Y. Chen L. Rilett, “Signal Timing Optimization for Corridors with
Multiple Highway-Rail Grade Crossings Using Genetic Algorithm,” [18] C. Canali, R. Lancellotti, “GASP: Genetic Algorithms for Service
Journal of Advanced Transportation, vol. 2018, ID. 9610430, pp. 1–14, Placement in Fog Computing Systems,” Algorithms, vol. 12, pp. 1–19,
2018. 2019.
[8] N. Saharkar , M. Wanjari, “A Genetic Algorithm Based Approach to [19] A. Rahman, N. Shahruddin, Ismail Ishak, “Solving the Goods
Solve Transport Problems for School Buses?” Journal of Engineering Transportation Problem Using Genetic Algorithm with Nearest-Node
and Applied Sciences, vol. 13, no. 4, pp. 848-851, 2018. Pairing Crossover Operator, ” Journal of Physics: Conference Series,
vol. 1366, pp. 1–9, 2019.
[9] T. Karthy and K .Ganesan, “Multi Objective Transportation Problem -
Genetic Algorithm Approach,” International Journal of Pure and [20] X. Feng, X. Zhu, X. Qian, Y. Jie, F. Ma, and X. Niu, “A new transit
Applied Mathematics, vol. 119, no. 9, pp. 343-350, 2018. network design study in consideration of transfer time composition,”
Transportation Research Part D: Transport and Environment, vol. 66,
[10] E. Han, H. PiLee, S. Park, J. So, I. Yun, “Optimal Signal Control pp. 85–94, 2019.
Algorithm for Signalized Intersections under a V2I Communication
Environment,” Atmoshphere, vol. 2019, ID. 6039741, pp. 1–9, 2019.

288 | P a g e
www.ijacsa.thesai.org