48 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 23, NO.

1, JANUARY 2022

A Review on Swarm Intelligence and Evolutionary

Algorithms for Solving the Traffic Signal
Control Problem
Palwasha W. Shaikh , Mohammed El-Abd , Senior Member, IEEE,
Mounib Khanafer, Senior Member, IEEE, and Kaizhou Gao

Abstract— The rapid development of urban cities coupled with

the rise in population has led to an exponentially growing number
of vehicles on the roads for the latter to commute. This is adding
to the already overbearing problem of traffic congestion. Short
term, costly and short-sighted solutions of road infrastructure
expansions are no longer suitable. One effective method of road
resource allocation is focusing on the widely used traffic signal
controllers’ timing schedules. Searching for a suitable or an
optimal schedule for the prior via brute force to ease traffic
congestion might not be the most elegant or feasible solution.
Nature-inspired algorithms including evolutionary and swarm
intelligence algorithms are gaining a lot of momentum. Many of
these algorithms have been used in the last two decades to address
different applications in the smart city era including traffic signal
control (TSC). This paper conducts a comprehensive literature
review on applications of evolutionary and swarm intelligence Fig. 1. The number of passenger cars and commercial vehicles in use
algorithms to TSC. Surveyed work is categorized based on the set worldwide between 2006 and 2015.
of decision variables, optimization objective(s), problem modeling
and solution encoding. The paper, based on gaps identified by the and commercial vehicles in use worldwide between 2006 and
conducted review, identifies promising future research directions 2015. In 2015 alone, around 947 million passenger cars
and discusses where the future research is headed. and 335 million commercial vehicles were in operation on
Index Terms— Evolutionary computation, evolutionary roadways worldwide [2], [3].
algorithm, swarm Intelligence, meta-heuristics, optimization, Recently, the INRIX Global Traffic [4], which is an impor-
traffic signal control, traffic intersection, single-objective, tant study about road traffic, revealed that U.S. alone has lost
multi-objective, bi-level optimization.
$305 billion due to traffic congestion; Los Angeles drivers
I. I NTRODUCTION spend an average of 102 hours sitting idle in traffic. Traffic
congestion has become an overbearing problem worldwide.
A CCORDING to the United Nations Department of Eco-
nomic and Social Affairs (DESA) [1], more than 68%
of the world’s population is expected to move to urban
The traditional solutions of expanding roadways and installing
intersections are not only costly, but also appear as poor
city areas by 2050. People living in the urban cities are short-term remedies.
expected to commute using their own private vehicles or An urban traffic system primarily consists of intersec-
public transport. Fig. 1 shows the number of passenger cars tions and links. Traffic congestions occur frequently in this
system, which poses critical challenges. The alleviation of
Manuscript received November 21, 2019; revised May 19, 2020; accepted these challenges not only improves commuting experience, but
July 15, 2020. Date of publication August 18, 2020; date of current version also mitigates several problems like environmental pollution.
December 28, 2021. This work was supported by a research grant from the
American University of Kuwait (AUK). The Associate Editor for this article In general, an urban traffic system is a heterogeneous system
was P. Ye. (Corresponding author: Mohammed El-Abd.) that has two types of intersections, namely, signalized and
Palwasha W. Shaikh is with the School of Electrical Engineering and non-signalized. As the names imply, a signalized intersection
Computer Science, University of Ottawa, Ottawa, ON K1N 6N5, Canada
(e-mail: [email protected]). utilizes traffic signals while a non-signalized intersection, also
Mohammed El-Abd and Mounib Khanafer are with the Department of known as an uncontrolled intersection, does not. In practice,
Engineering, American University of Kuwait, Safat 13034, Kuwait (e-mail: both types of intersections are adopted in an urban traffic
[email protected]; [email protected]).
Kaizhou Gao is with the Institute of Systems Engineering, Macau University system, and this leads to the heterogeneous features of link
of Science and Technology, Taipa 999078, Macau SAR, China, and also with volume dynamics and traffic flow. In [5], the authors conducted
the Collaborative Laboratory for Intelligent Science and Systems, Macau a detailed survey on systems for monitoring intersections and
University of Science and Technology, Taipa 999078, Macau SAR, China
(e-mail: [email protected]). studied the behaviors of road users (vehicles, drivers, and
Digital Object Identifier 10.1109/TITS.2020.3014296 pedestrians). The authors also provided a microscopic view
1558-0016 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: SASI INSTITUTE OF TECHNOLOGY & ENGINEERING. Downloaded on October 23,2024 at 05:35:24 UTC from IEEE Xplore. Restrictions apply.
 SHAIKH et al.: REVIEW ON SWARM INTELLIGENCE AND EVOLUTIONARY ALGORITHMS 49

of the traffic dynamics at the intersections. For a macroscopic

view of a traffic system, the studies in [6]–[13] investigated
non-signalized intersections.
Previously, a Traffic Signal Control (TSC) problem was
solved by using a fixed-time cycle strategy, with individ-
ual or coordinated intersections to adjust the green sig-
nal ratio in a traffic system [14], [15]. For traffic light
design, there are four different traffic signal control strate-
gies, namely, fixed-time versus traffic responsive, and isolated
versus coordinated. Notable classic strategies proposed in the
last few decades include, MAXBAND [16], TRANSYT [17],
SCOOT [18], OPAC [19], PRODYN [20], CRONOS [21],
and RHODES [22]. From a system modelling perspective,
Lighthill and Whitham in 1955 and Richards in 1956 provided Fig. 2. The number of indexed publications covering TSC optimization using
a good foundation in macroscopic traffic system modelling; evolutionary computation in different databases from 2010 to 2019.
thus, yielding the LWR model [23], [24]. A breakthrough
came when Daganzo [25], [26] developed a finite difference algorithms could be easily applied to complex optimization
solution scheme for the LWR model by adopting a simplified problems. Furthermore, in many large-scale engineering prob-
fundamental diagram called a cell-transmission model (CTM). lems, especially those that involve uncertainties like the traffic
In [27], [28], CTM was transformed into a mixed integer linear flow problem, obtaining a close-to-optimal solution in reason-
programming (MILP) problem. Following CTM, time Petri able time is satisfactory enough. Metaheuristics can provide
nets were employed in [29] to model and control urban road high-quality near-optimal solutions with an affordable compu-
intersections with incidents. tational cost as opposed to conventional optimization meth-
The term population-based metaheuristics refers to a class ods [43], [44]. These population-based algorithms have been
of algorithms used to tackle non-linear continuous and/or successfully applied to many applications including, but not
discrete optimization problems. Such algorithms store a pop- limited to, industrial applications [45], feature selection [46],
ulation of individuals (i.e., solutions) and update these indi- intrusion detection [47], dynamic optimization [48], economic
viduals for several iterations (i.e., generations) until a certain dispatch [49], and truck scheduling [50], [51].
stopping criterion is satisfied. This class of algorithms could be The TSC problem is tackled in the literature by different
further divided based on the inspiration behind their solution population-based algorithms. Based on a search conducted
update technique. The first type is Evolutionary Algorithms using different search engines, Fig. 2 illustrates the number
(EAs) [30]–[32]. In EAs, the update process is inspired by of research papers that have been published over a span
the Darwinian theory of evolution [33]. These algorithms of 10 years (from 2010 to 2019) to address the TSC problem
include Genetic Algorithms (GAs), Genetic Programming using such methods. The increasing pattern is reflective of the
(GP), Evolutionary Strategies (ES), Evolutionary Program- growing interest in the topic and highlights the need for the
ming (EP), and Differential Evolution (DE). The second type current work.
is Swarm Intelligence (SI) algorithms. In SI, the update Previous surveys in the literature covered the tackling
process is inspired by an observed behavior of some living of TSC using Computational Intelligence (CI) [52]–[54],
entity (humans, animals, insects …etc.). The term foraging Reinforcement Learning (RL) [55], [56], and Deep RL meth-
algorithms is used to describe SI algorithms that mimic the ods [57]. The work in [58] highlighted that in the past decade,
foraging behavior of certain species. Examples of such algo- Artificial Intelligence (AI) methods including, among others,
rithms include, among others, Particle Swarm Optimization group intelligence algorithms (i.e. metaheuristics), dominate
(PSO) [34], [35], Ant Colony Optimization (ACO) [36], and the many traffic control methods based on intelligent comput-
Artificial Bee Colony (ABC) [37]. Other SI algorithms are ing. However, and to the best of our knowledge, there has not
inspired by different kinds of behaviors including, for example, been a literature review covering the application of EA and SI
the egg laying behavior of cuckoos in Cuckoo Search (CS) algorithms in TSC. Although the work in [59] covers a similar
[38], the echolocation behavior of bats in the Bat Algo- topic, the current work provides more contributions as follows:
rithm (BA) [39], the brainstorming behavior in the Brain Storm • Conducting a comprehensive literature review on opti-
Optimization (BSO) algorithm [40], and musical ensembles in mizing the TSC problem using EA and SI approaches.
Harmony Search (HS) [41]. • Comparing and categorizing the surveyed literature based
Metaheuristics in general do not require any prior infor- on the set of decision variables, optimization objective(s),
mation about the search space characteristics [42]. This problem modeling and solution encoding.
property makes such algorithms very desirable in dealing • Providing insights into future research directions based
with complicated problems having problematic characteris- on identified gaps in the current research.
tics including, but not limited to, discontinuity and non- The rest of the paper is organized as follows. Section II
differentiability. Moreover, metaheuristics usually entail less identifies the objectives and parameters usually used in the
complexity in comparison with exact methods that require the development of optimization models for the TSC problem.
first-order and second-order information. Hence, metaheuristic Section III provides a brief description of population-based

Authorized licensed use limited to: SASI INSTITUTE OF TECHNOLOGY & ENGINEERING. Downloaded on October 23,2024 at 05:35:24 UTC from IEEE Xplore. Restrictions apply.
 50 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 23, NO. 1, JANUARY 2022

as optimizing the vehicle flow of either public or private

vehicles or both. Optimization foci could range from high
throughput rate to reduced queue and stop-and-wait times.
EAs and SI are used most often to achieve these goals.
• Optimizing the Intersection: TSC is often implemented
to improve the traffic conditions at the intersection. Traffic
states may include the arrival rate of the vehicle, length
of the queue on each road of the intersection, wait times
…etc. These indicators are acquired usually via sensors in
real-time. Single-intersection or multi-intersection were
studied under signal cooperative control. The approach
Fig. 3. A traffic intersection with four phases. of studying these intersection optimization problems in
isolation is motivated by the observation that the traffic
metaheuristics. Section IV surveys previous EA applications problem may appear first at one intersection and then
to TSC. Section V surveys previous SI applications to TSC. begin to spread to others. Take, for instance, the problem
Comparative studies are presented in Section VI. Finally, of the queue left from the previous cycle trying to
Section VII concludes this current work and outlines future disrupt the important movement of the next cycle (a
research directions. problem known as spillback). Dealing with the prior at
the isolated emergence point may ease the traffic flow.
II. P ROBLEM D EFINITION Even the residual queues could be avoided using the same
A. Traffic Networks approach. These local traffic parameters are influenced by
the intersection’s environment.
A traffic network consists of a set of road links connected
• Optimizing the Roundabouts: Often modelled as a
via intersections (or junctions). Each intersection consists of a
concatenation of junctions, roundabouts are constructed
number of approaches and the crossing area. An approach may
to increase traffic safety and traffic capacity. As a result,
have one or more lanes but has a unique, independent queue.
roundabouts often replace intersections. Optimization of
Approaches are used by corresponding traffic streams. Two
roundabouts aims to reduce time delays, cycle times,
compatible streams can safely cross the intersection simulta-
vehicle delays and expenses, even though they may
neously, while antagonistic streams cannot. In traditional TSC,
irk drivers and become inefficient under heavy traffic
a signal cycle is one repetition of the basic series of stages
flows. Hence, using a TSC may optimize roundabouts
at an intersection, where each stage consists of simultaneous
effectively.
traffic light signals allowing a predefined compatible traffic
• Optimizing Lights or Timing Schedule Cycles: Traffic
streams to cross the intersection simultaneously. The duration
lights have red, green and yellow phases. Depending on
of a cycle is called cycle time. For safety reasons, constant lost
the location and traffic flow, the duration of the phases
(or intergreen) times of a few seconds are necessarily inserted
varies; the red phase may last longer or shorter than
between consecutive stages to avoid interference between
the green phase, for example. Traffic flow is reliant on
antagonistic streams. For each traffic light, the ratio of the
these traffic light schedules, as drivers interact with their
green and red times within one cycle is called split, and a delay
physical environment. More often and to the driver’s
between the starting times of green periods of two neighboring
frustration, the red phases of traffic lights will interrupt
traffic lights along the same traffic route is called offset.
the smooth flow of traffic and create queues. If there is
A sample general intersection with four directions is shown
no congestion, these queues dissipate during the green
in Fig. 3. The illustrated intersection has four phases for
interval. However, under situations of congestion, red
vehicles and pedestrians.
phases only lead to additional delays. Thus, signal timings
impact the queue length of each phase. Determining the
B. TSC Goals and Parameters green phase is not an easy task. EA and SI often converge
Since the economic and ecological costs of physically to a near-optimal solution. Other possible parameters to
expanding the road infrastructure are too high, the need to be optimized include wait and travel times, urban traffic
divert our attention to other aspects of the road network arises. pressure and clearance time for the queue of vehicles.
TSC relates parameters like wait times and queue length with Fig. 4 summarizes this section; it presents the division of
traffic signal parameters, such as cycle length, green time, the TSC problem according to the most targeted objectives
split, and offset. Hence, optimizing traffic flow on existing and the respective parameters manipulated to achieve those
roads is often targeted in parallel with optimizing the traffic objectives.
signal control. The following subsections cover the different
optimization aspects that fall under TSC:
III. P OPULATION -BASED M ETAHEURISTICS
• Optimizing the Network: Optimizing road networks is
AND T HEIR A PPLICATION F OR TSC
not just limited to expanding roads and highways, it can
also refer to smoothing traffic flows by inserting TSC This section gives a brief description on population-based
operations at an intersection. It could even be as simple metaheuristics and how they are applied to the TSC problem.

Authorized licensed use limited to: SASI INSTITUTE OF TECHNOLOGY & ENGINEERING. Downloaded on October 23,2024 at 05:35:24 UTC from IEEE Xplore. Restrictions apply.
 SHAIKH et al.: REVIEW ON SWARM INTELLIGENCE AND EVOLUTIONARY ALGORITHMS 51

Fig. 4. Division of the TSC problem based on the object ives and their most commonly manipulated parameters.

A. Population-Based Metaheuristics

As stated in the introduction, population-based metaheuris-

tics are divided into EAs and SI. However, both types of
algorithms have a similar framework for optimizing a given
problem. First, all individuals (i.e. candidate solutions) in the
population are initialized with some solutions generated ran-
domly (uniformly or otherwise) or by simple greedy heuristics,
like dispatch rules. The values of all decision variables should
be in their defined search ranges or domains. Then, the initial
solutions are evaluated using the problem defined objective
function. Next, an iterative process is repeated to generate
new solutions using different nature-inspired mechanisms.
In EAs, such mechanisms include the steps of selection,
crossover, and mutation inspired from the Darwinian theory
of evolution. On the other hand, in SI such mechanisms are
derived from the equations of motion of the living entity
behind the algorithm’s inspiration. Examples of these models
include the concept of positions and velocities in PSO and
the pheromone update rules in ACO. The new individuals are
evaluated to replace the old ones in the population based on
set rules (e.g. solution replaced if improved) that are often
set in advance. The objective values of all individuals in
the population are updated as well. Moreover, some local
search or randomization techniques might be invoked during
the search to improve the exploitation (e.g. extensively search Fig. 5. A general flowchart for population-based metaheuristics.
within a specific region) or exploration (e.g. explore new areas
in the search space) behaviors. The iterative process is repeated B. Application for TSC
until some stopping criterion is met (e.g. a specified number
of iterations, a specified number of function evaluations, To solve TSC, many meta-heuristics are employed,
an acceptable error threshold, a time limit …etc.). Finally, improved, or hybridized with other techniques for an enhanced
the best solution(s) along with its objective value(s) is provided performance. First, a mathematical model relating the opti-
by the algorithm. A general flowchart for population-based mized objectives (e.g. wait times, queue length. . .etc.) to
metaheuristics is shown in Fig. 5. the decision variables (e.g. green time, split …etc.) must be

Authorized licensed use limited to: SASI INSTITUTE OF TECHNOLOGY & ENGINEERING. Downloaded on October 23,2024 at 05:35:24 UTC from IEEE Xplore. Restrictions apply.
 52 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 23, NO. 1, JANUARY 2022

developed. In case of tackling multiple objective, the prob-

lem could be addressed as a single objective optimization
problem (SO) by using weighted average or as a purely
multi-objective optimization problem (MO). Some models use
bi-level optimization in which traffic signal parameters are
usually handled at the upper level. Once the problem is prop-
erly formulated and decision variables are selected, solution
encoding is essential to represent individuals in the population.
Binary, integer, and real-coded representations have all been
attempted in the literature. Based on the selected encoding
scheme, population update mechanisms might be slightly
updated and/or modified. In order to evaluate the objective
function of an individual, the signal plan encoded inside the
individual is simulated using a simulation tool (e.g. VISSIM)
in order to find the values for the selected objective. Finally, Fig. 6. Percentage of publications on TSC optimization problem in the period
the population is updated, and this iterative process continues 2001-2019.
until a stopping criterion is reached. Larger traffic networks
(e.g. with increasing number of intersections), increases the A. Genetic Algorithms in TSC
size of the individuals as it needs to contain all parameters The work in [60] proposed the use of GA combined
for all traffic signals. This raises the issue of computa- with urban traffic microscopic simulation model (UTMSM)
tional complexity and the applicability of such algorithms in and international vehicle emissions (IEV) model to reduce
real-time. the maximize the intersection capacity and reduce vehicle
Manual data extraction was adopted in order to conduct emission. This was a MO function formulated as a SO using
search for the current survey. The manual search is performed weights and normalization to account for different units and
on the most relevant sources, namely: IEEEXplore, ElSevier, order of magnitude. The solution was formulated as a real-
Springer, Wiley, and ACM. The selected databases index all coded vector representing the green time of each phase of the
kinds of publications including conference proceeding, journal intersection. The approach was applied to a single intersection
papers, articles and book chapters. The same search query was in Wuhan, China. The algorithm provided an overall 9.8% in
run on all databases containing specific keywords for the time average vehicle speed and a 3.5% reduction in emission in
period 2001 to 2019. The query was divided into two ANDed comparison to the original plan.
terms, one term related to the problem domain (i.e. applica- GA was applied in [61] to handle a large-scale real-world
tion) and one term related to the class of adopted algorithmic case in a congested situation using hypothetical test cases
solutions. The first term was devised to extract publications with the use of cellular-automata-based micro-simulators and
related to “Traffic Signal”, “Traffic Signal Control”, “Traffic a Beowulf cluster. Test cases where generated based on con-
Intersection”, or “Roundabouts”. The second term contained gested scenarios in Saragossa, Spain. Four different objectives
both generic keywords (related to the algorithmic class in were tested individually as a SO: the number of vehicles (NoV)
general) and specific keywords (related to the most prominent leaving the network, the mean travel time (MTT), the state
algorithms). It was devised to extract publications related of occupancy (SOC), and the global mean speed. The aim
to “Evolutionary Algorithm”, “Swarm Intelligence”, “Evo- behind this was to detect any correlation among the evolution
lutionary Computation”, “Metaheuristic”, “Particle Swarm of each variable. The solution was represented as a stage
Optimization”, “Genetic Algorithm”, “Artificial Bee Colony”, sequence for each intersection encoded using a binary gray-
“Differential Evolution”, or “Ant Colony”. Raw literature data code. The authors discovered a strong linear relationship
was filtered to extract publications that specifically manipulate between the MTT and SOC and a slight relation between SOC
traffic signal schedules/timings, present reproducible exper- and NoV. The work proposed using a multi-criterion objective
iments, and published during the last decade in reputable function.
venues. The work in [62] addresses the traffic signal timing man-
Fig. 6 presents a pie-chart, created based on data extracted agement problem using GA. Integer encoding was used for
from the different databases, showing the percentage of the solution, which contained the offset, the cycle time,
publications that tackled the TSC optimization problem, the phase sequence, and the green signal split. The aim
based on different population-based metaheuristics, in the was to optimize the traffic delay and fluency. The problem
period 2001-2019. was modelled by considering each solution as an agent pre-
sented with an incoming customer (i.e. vehicle). The agent
IV. S URVEY OF E VOLUTIONARY is awarded when the vehicle reaches the destination based
A LGORITHMS FOR TSC on the travelled distance and the waiting time. The perfor-
This section surveys previous works in the literature that mance of the algorithm was evaluated in a two-intersections
applied EAs to the TSC problem. A comparative summary network given two scenarios of undersaturated and oversat-
is provided for applications of GAs in Tables I and DE urated networks. The algorithm was able to avoid queue
in Table II. spillback.

Authorized licensed use limited to: SASI INSTITUTE OF TECHNOLOGY & ENGINEERING. Downloaded on October 23,2024 at 05:35:24 UTC from IEEE Xplore. Restrictions apply.
 SHAIKH et al.: REVIEW ON SWARM INTELLIGENCE AND EVOLUTIONARY ALGORITHMS 53

TABLE I
R EPORTED GA APPROACHES FOR THE TSC PROBLEM

In [63], the authors applied NSGA-II [64] to solve the traffic intersections, with each component containing the cycle time,
signaling optimization problem with two objectives. The two the offset, and the green time of each phase. Experimental
objectives considered were the average delay time and average results illustrated a good performance for their implementation
stop times. In order to provide detailed micro-simulations, and with a speedup above 20 for both cases.
to account for the huge computational burden, the algorithm NSGA-II was also adopted in [65] to optimize oversaturated
is implemented on a Graphic Processing Unit (GPU) and run traffic signal intersections. The goal was to maximize the
on a 4 and 18 intersection networks in Zhongguancun, China. throughput and minimize the queue length. The solution was
Binary encoding was used for solution representation. A single encoded as real numbers representing the green time of each
individual had I components, where I refers to the number of phase. Simulations run on a 4-way intersection illustrated that

Authorized licensed use limited to: SASI INSTITUTE OF TECHNOLOGY & ENGINEERING. Downloaded on October 23,2024 at 05:35:24 UTC from IEEE Xplore. Restrictions apply.
 54 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 23, NO. 1, JANUARY 2022

TABLE II
R EPORTED DE APPROACHES FOR THE TSC PROBLEM

the signal timing plan generated good efficiency in oversatu- condition was reduced to 52.2 sec/veh or equivalent to level of
rated conditions with respect to the commonly utilized signal service (LOS) D.
timing optimization software Synchro. A bi-level optimization model for an eco-traffic signal
In [66], the authors proposed an improved real-coded GA system was proposed in [71]. In the developed model, the
to optimize an actual intersection Shenzhen, China. GA opti- upper-level (an eco-signal operation) was responsible for
mized the traffic emission, average delay, and traffic capacity. optimizing the total delay by manipulating the timing plan.
The solution contained the green time for each traffic light. On other hand, the lower-level (an eco-driving algorithm)
The proposed algorithm provided better fitness and faster was to optimize the total fuel consumption by controlling the
convergence than the original GA. It also improved the delays, vehicles’ speeds and accelerations. The Bi-level GA (BiGA)
emissions and capacity with about 1% to 15% improvement in [72] was used to optimize the upper-level problem. Exper-
over the classical Webster (TRRL) timing scheme. iments run on a hypothetical signalized intersection showed
GA was used in [67] to simultaneously find best routes 5-10% fuel savings and 12% travel-time reduction over actu-
for vehicles and optimize traffic signal timings in real time. ated signal control and eco-driving algorithm.
This was an extension over the work in [68], which only The authors in [73] developed a framework for real-time
optimized vehicle routes. The solution was modelled as an arterial coordination with three sub-models. The framework
array having a length equal t the number of vehicles plus the was based on dynamic intersection turning fraction estimation.
number of traffic lights. The gene related to a vehicle depicts Two improved GAs were developed to solve two of the sub-
the route number while the gene related to a traffic light models. The first sub-model was developed for the dynamic
represents the green time. The objective function used was turning fraction estimation. The solution was represented as
the total time taken to dispatch the vehicles based on a Petri- a matrix, with binary encoding, having r rows and (s − 2)
net simulation model. Experiments were run using a network columns, where r and s are entering and exiting links,
of 12 intersections and analysis showed that the adopted GA respectively. The objective function was to maximize the
provided good optimization by keeping vehicles on the road throughput. However, the third sub-model was developed to
for the smallest amount of time. optimize the offset and split based on the achieved public cycle
Comparison experiments between GA, as a global opti- length. A real-coded GA was developed for that model, which
mizer, and Expectation Minimization (EM), as local optimizer, maximizes the non-stop movement. An arterial road in Beijing,
were conducted in [69] using large-scale traffic networks. Each China was taken as a case study. The proposed framework
individual was a vector of structures, where each structure produced less queue lengths, less delay time (except for the
contains the information of each intersection, namely: the first cycle), and less number of stops than MAXBAND.
horizontal green time, the vertical green time, the state (i.e. To provide a real-time application, an improved GA was
which direction is active), and a time counter (i.e. remining developed in [74] to control a 4-phase 3-intersections network.
time for changing the current state). The fitness used was Optimization variables were selected as the green light times
the average traffic time for all vehicles to arrive at their and phase differences between intersections. Experimental
destinations. GA has showed a better performance, especially results showed that the developed algorithm required 38% less
with increasing number of vehicles, on the expense of having search time than a standard GA while reducing the average
higher memory consumption. delay time by 34.57%. it also provided a reduced average delay
The work in [70] adopts GA to minimize the vehicle delay of 17.3% with respect to an improved CTM.
in a signalized intersection under oversaturated conditions. In [75], another bi-level optimization model was proposed
The decision variable used was the effective green time for optimizing traffic signal settings. The upper-level was
ratio in every phase. A simulation environment, a single formulated to determine the traffic signal settings in order
intersection, was designed based on the guideline of Pub- to minimize average travel time. The lower-level problem’s
lic Works Department Malaysia (JKR). Simulations showed goal was to achieve the network equilibrium using the settings
that GA was able to provide appropriate green time for provided by the upper-level. A hybrid GA was used to convert
each phase based on the traffic demand. The overall aver- the problem into two single-level problems that are solved
age delay of the simulated intersection under oversaturated sequentially. The solution was represented as real numbers

Authorized licensed use limited to: SASI INSTITUTE OF TECHNOLOGY & ENGINEERING. Downloaded on October 23,2024 at 05:35:24 UTC from IEEE Xplore. Restrictions apply.
 SHAIKH et al.: REVIEW ON SWARM INTELLIGENCE AND EVOLUTIONARY ALGORITHMS 55

representing the offset, green, and red times of a signal. The model required around 45% and 16% less computational
implementation was also parallelized using a master-slave time in comparison with the GA- and HS-based models,
architecture to accelerate fitness calculations. Experiments respectively.
conducted on a synthetic traffic network and a real-world traf- DE is used in [85] to simultaneously optimize the signal
fic subnetwork showed that the proposed framework provided timing and signal phasing of a 3-ntersections network. DE was
smooth transitions between successive traffic signal settings used to minimize the average vehicle delay. Decision variables
by taking offsets into account. were the three signal times of three phases. The proposed DE
GA was improved, in [76], by incorporating a approach resulted in an average vehicle delay reduction in the
neighborhood-based simple descent algorithm as a local range of 28%–42% and 3%–38% compared to the optimum
search technique. A single solution included green times fixed-time signal control and vehicle-actuated signal control
of all phases at every intersection. The objective was to for the studied test cases, respectively.
minimize the average traffic delay. The algorithm was tested The work in [86] proposed a bi-level model to optimize
in two different cases in Brisbane, Australia, and Plock, environmentally friendly traffic signal timings. The upper-
Poland. Results showed that the algorithm outperformed GA level was modeled as MOP maximizing the reserve capacity
(without local search) and traditional fixed-time traffic signal of the network and minimizing the vehicles’ emissions. The
settings. lower-level was designed to evaluate the users’ reactions.
In [77], a multi-objective model was proposed to address DE was applied at the higher level Both levels were addressed
the signal timing problem. Green times of each phase was with DE in which the solution is represented as both real-
optimized to minimize emission, total delay, and the number valued and integer variables representing the O-D (Origin-
of stops. GA was used to find the optimal signal plan. The Destination) multipliers, the cycle times, and the green times.
proposed optimization model with GA achieved better results The framework was applied to two networks having two-
compared to the field conditions. Moreover, although the total signalized and 9-signalized intersections. Pareto-front analysis
stops increased by 4.35–5.22%, the total delay has decreased revealed the usefulness of the model to provide opportunities
by 7.86–12.71%. In addition, there was a noticeable reduction for authorities to be able to balance between two conflicting
in all types of emissions. objectives.
The work in [78] adopted a multi-objective GA to optimize
traffic in a 9-intersections network. The algorithm ran off-line
C. Genetic Programming in TSC
based on prior traffic information. The algorithm handled four
objectives in order to maximize the throughput, minimize the The works in [87], [88] reported the first application of
travelling delay, enhance traffic safety, and avoid spillovers. GP in TSC. Instead of designing a single general controller,
It was concluded that the size of the traffic network and their approach evolved a controller for each intersection to
its geometries affect the computation time of the algorithm. be optimized. The evolved solution was modeled as a forest
Moreover, the proposed control designs were able to deal of syntactic trees, where each tree represents an indepen-
small traffic disturbances. The work was also extended in [79] dent program for traffic signal control. The objective was to
to handle one additional objective, which is the intersection minimize the average vehicle delay. The results indicate a
crossing volume. significant improvement over static and actuated traffic signals
In [80], an improved GA was used to solve the TSC problem in scenarios with medium- and high-density traffic.
considering travel safety of the elderly. The algorithm was
adopted to optimize the average vehicle delay by controlling V. S URVEY OF S WARM I NTELLIGENCE
the green signal ratio and cycle time. The algorithm was A LGORITHMS FOR TSC
applied to the intersection of Wenfeng road and Jiaochang road This section surveys previous works in the literature that
in Lintao County of Gansu Province, China. The proposed applied SI algorithms to the TSC problem. A comparative
algorithm provided a 24% improvement on average compared summary is provided for applications of PSO in Tables III,
to the Webster algorithm. ACO in Table IV and HS in Table V.

B. Differential Evolution in TSC A. Particle Swarm Optimization in TSC

The work in [81] formulated the design of a signal- A simulated annealing PSO (Sa-PSO) was developed ad
controlled road networks as a bi-level optimization problem. applied to urban traffic signal timing in [89]. Decision vari-
The upper-level minimizes the total network performance ables included the green time and cycle times. Simulations
index while the lower-level models a traffic assignment prob- were conducted on a 9-intersection network and the result
lem representing route selections of users. DE was applied to illustrated that Sa-PSO was able to reduce the average delay
the upper-level. The solution was composed of the network per vehicles by 41.0% and the average stop rates by 30.6% in
cycle time, intersection green times, and offset variables. The comparison to fixed time plans.
framework was applied to a model test road network [82] A PSO algorithm was employed in [90] to find the opti-
and results illustrated improvements over results obtained by mized cycle programs (OCP) for all traffic lights located in
GA-[83] and HS-based models [84] in terms of the total an urban area. This OCP was encoded as a vector of integers,
network performance index. Moreover, the proposed DE-based where each element represents a phase duration of one state

Authorized licensed use limited to: SASI INSTITUTE OF TECHNOLOGY & ENGINEERING. Downloaded on October 23,2024 at 05:35:24 UTC from IEEE Xplore. Restrictions apply.
 56 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 23, NO. 1, JANUARY 2022

TABLE III
R EPORTED PSO APPROACHES FOR THE TSC PROBLEM

TABLE IV
R EPORTED ACO APPROACHES FOR THE TSC PROBLEM

TABLE V
R EPORTED HS APPROACHES FOR THE TSC PROBLEM

Authorized licensed use limited to: SASI INSTITUTE OF TECHNOLOGY & ENGINEERING. Downloaded on October 23,2024 at 05:35:24 UTC from IEEE Xplore. Restrictions apply.
 SHAIKH et al.: REVIEW ON SWARM INTELLIGENCE AND EVOLUTIONARY ALGORITHMS 57

of the traffic lights. The problem was a SO problem, where in China. The proposed algorithm exhibited a faster conver-
the fitness function is formulated to maximize the number of gence rate and the less total delay than the GA proposed
vehicles reaching their destination (placing it in the denom- in [98].
inator) and to minimize the global trip time of all vehicles To handle dynamic and real-time optimization scheduling of
(placing it in the numerator) during the simulation time. urban traffic lights, the work in [99] proposed inner and outer
The algorithm was applied to two large and heterogeneous cellular automation models (ICM and OCM) combined with
metropolitan areas located in Málaga and Sevilla in Spain. PSO (IOCA-PSO). ICM was responsible for the phase shifting
Results reported an improvement of 0.13% in completed trips control of traffic lights while PSO helps the OCM effectively
and 15.7% in trip time when compared to a deterministic find the optimal timing scheduling in ICM in limited time.
algorithm. The authors extended the same work in [91] by PSO was used to minimize the percentage of waiting time
comparing against a standard PSO algorithm and DE where and red-light time by modifying the time span matrix of
their algorithm produced statistically better results. ICM. Since each particle is a matrix, movement equations
The work in [92] applied an improved PSO algorithm to are modified such that the particle, to be updated, selects the
TSC of a 4-way intersection built based on real data. The particle with the lowest fitness value in its neighborhood and
algorithm optimized the average vehicle travel time by manip- jumps to it. If a particle selected itself, it would not be updated
ulating the green time ratio. The proposed algorithm provided in the current iteration. As a test case, an area in Wuhan was
a reduction of 24.1% in average running time compared to a used as the objective urban area. The algorithm was compared
traditional timing scheme. to GA and PSO methods. Comparison results revealed that
In [93], PSO was applied to control a real road network IOCA-PSO achieved the best performance in small number of
that consists of four junctions/nodes having different time iterations. Moreover, IOCA-PSO was shown to be capable of
signaling models. The network studied was the Ooe Toroku handling more vehicles than the other methods before traffic
road network of Kumamoto Shi, Japan. The aim was to jams were built.
optimize vehicle wait out, vehicle in, travel distance, and time The work in [100] proposed a pareto-front multi-objective
delay. Decision variables were the traffic light signal parame- PSO for real-time traffic signal control. The goal was to
ters consisting of the offset, cycles, and splits time of each minimize delay time, minimize number of stops, and maximize
node of the considered road networks. Experimental results effective capacity. The algorithm manipulated the signal time
showed that PSO provided better performance the Multi- of each phase to optimize the objective functions. The test
Element Genetic Algorithm (ME-GA) [94] by increasing the case selected was at the intersection of Zhao Dengyu Road
percentage of vehicle flow by about 4.13%. and Pinganli West Street in Beijing, China. The algorithm
A bi-level model for traffic network signal control was pro- was shown to be applicable to time-varying travel demand
posed in [95]. The lower-level was a dynamic user equilibrium and robust enough for on-line applications.
with embedded dynamic network loading. The upper-level In [101], an enhanced PSO was proposed to minimize the
manipulated the signal green splits to minimize the network- queue lengths in a traffic system by optimizing the traffic
wide travel cost. The model was applied to test networks signal plan. The algorithm considered current queue length,
and solved using three different metaheuristics. For a seven- the recorded traffic flow, the delay at each phase, and the
arc network with constant signal splits, PSO has achieved a average waiting time in each direction to suggest the traffic
relative optimality gap of 0.19%. For a seven-arc network right-of-way assignment for each traffic phase. Compared to
with variant signal splits, PSO substantially outperformed fixed control, PSO proved to be more efficient in dealing with
Simulated Annealing (SA). Moreover, the authors reported a high traffic flow as it significantly reduced the queue length
parallel implementation of the algorithm on 4 computing nodes of the congested direction.
and reported an improvement of 61.05% in the computational Adaptive traffic control was tackled using a hybrid PSO
efficiency. and Tabu search algorithm in [102]. In their implementation,
Arterial traffic signal timing parameters were optimized good solutions found by PSO were kept in a Tabu list to
by PSO in [96]. PSO was employed to find the optimal avoid getting trapped in local minima. Each particle con-
values for the effective green times and offsets on coordinated tained the green timings of all links across all intersections
routes. The objective was to minimize the total delay. The with the objective of minimizing the total delay. Bab Zaer
algorithm was tested on an arterial with three intersections intersection in Rabat City, Morocco was taken as a test case.
located in Prices Fork Road, Blacksburg, Virginia. Timing A classical PSO algorithm was able to reduce the delay by at
planes generated by PSO reduced the delay by 13%, 24% and least 79% compared to the static plan. Moreover, the hybrid
44% in three different scenarios in comparison with timing PSO-TS algorithm reached the same solution faster than
plans generated by Vistro. The authors highlighted the need PSO.
to alleviate the computational time by boosting the computing The authors in [103] proposed a dynamic chaotic and self-
device. adaptive PSO to minimize the delay of each passenger at
In [97], the authors modeled the online traffic network using a single intersection. The aim was the application of this
a CTM and applied PSO to optimize the control of the traffic control method for public transit priority traffic signal. Each
signals. Each particle contained the signal plan for all junction particle is composed of the green time of each phase in a
and the objective was to minimize the total dely. The algorithm 4-phase intersection. Chaotic logistic model was used in the
was applied to an urban sub-network in the city of Changsha particles’ initialization and a self-adaptive strategy was used

Authorized licensed use limited to: SASI INSTITUTE OF TECHNOLOGY & ENGINEERING. Downloaded on October 23,2024 at 05:35:24 UTC from IEEE Xplore. Restrictions apply.
 58 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 23, NO. 1, JANUARY 2022

to set the inertia weight of PSO. The algorithm provided less In [107], the deployment of the rank-based ant system to
average delays for social vehicles, public transits, and pas- solve the TSC problem was investigated. The authors improved
sengers when compared to the Webster method and a classic the algorithm by adding a local search technique employed a
PSO. heuristic value in the form of an exponential function into
A multi-objective PSO algorithm was proposed in [104] to the transition probability of the ACO algorithm. The algo-
optimize per capita delay, vehicle emissions, and intersection rithms were evaluated on a four-legged intersection point. The
capacity. Each particle contained the signal cycle and green objective was to minimize the average vehicle waiting time at
times. The algorithm was applied to a simple four-way inter- the intersection point. The improved rank-based ant system
section in Jinzhou, China. The generated pareto-front solutions was compared to a fully actuated control algorithm based on
had better uniformity and diversity when compared to what is the National Electrical Manufacturers Association (NEMA)
generated by NSGA-II. Moreover, the computational time of standard through simulations. It was found that for high arrival
the proposed algorithm was superior NSGA-II under the same rates, the proposed ACO algorithm managed to improve the
hardware environment. average vehicle waiting time by 85%.
The problem of transition between signal timing plans
B. Ant Colony Optimization in TSC during the day was addressed in [108]. This transition is
The authors in [105] studied the use of ACO and SGA to required due to the continuous changes in traffic patterns. The
tackle the oversaturated traffic network problem. The objective authors proposed a mathematical model to better coordinate
was to maximize the number of vehicles released by a traffic signals at intersections during the transition phase. The math-
signal (that is, the throughput). To evaluate the performance of ematical model was based on a single-objective optimization
ACO and SGA algorithms, two network models were utilized. problem that aims at minimizing the social cost during the
The first model was a hypothetical model based on a one- transition period. The social cost is expressed as a function
way traffic network with 20 signals. The second model was of the delay cost, fuel consumption cost, and air emission
a realistic one that is based on the downtown traffic signal cost. In solving the optimization problem, the ACO algorithm
network of the city of Fort Worth, Texas. The network was is utilized and a hypothetical network with three signalized
composed of a 4 × 4 traffic signal. The performance ACO intersections is considered. Simulations were conducted on
and SGA was analyzed using MATLAB. For the first network AIMSUN to compare the transition timing plan obtained by
model, ACO outperformed SGA and managed to reach a applying the ACO algorithm with transitions based on tradi-
solution that allows for 3000 vehicles to be processed through tional methods, namely, immediate, two-cycle, and three-cycle
the network, while minimizing the ideal offset (that is, the time transition methods. Simulations revealed that ACO algorithm
difference between green times of two adjacent signals) to an improved the social cost by an average of 25%. This entailed
average of 2 seconds for each cycle and at each intersection in better vehicle operating costs and gas emission costs by almost
the network. For the second network model, the SGA managed 20%, better delay by almost 5%, and better time value cot by
to specify more fit solutions with low computational power. almost 22%. Simulations also revealed that the proposed ACO
However, ACO proved to be more reliable as it caused less algorithm achieved superior performance when applied to
variability in the set of results. ACO identified signal timings main streets. However, when secondary streets are considered,
that allowed for 4000 vehicles to be processed through the the three-cycle method outperforms the ACO algorithm in all
network and minimized the ideal offset to 1 second for each performance parameters.
cycle at each intersection.
The authors in [106] pointed out that in determining optimal
signal timings, the behaviour of the drivers in choosing routes C. Harmony Search in TSC
in the road network should be considered. As such, the authors In [109], a bi-level model is proposed for solving the
approach the TSC problem by considering a bi-level structure Equilibrium Network Design Problem (ENDP). HS was used
in which optimizing the signal timings is handled at the upper at the upper-level to optimize the signal timings, while the
level while traffic assignment is handled at the lower level. traffic assignment was handled at the lower-level. Decision
For the upper level problem, the authors proposed to use an variables handled by HS included the cycle length, the green
improved ACO algorithm to produce better signal timings. times and the offsets. the objective was to minimize the
The authors designed a single-objective optimization problem delay and number of stops for each link. The algorithm was
to minimize the total cost of a given road network. The applied to a network having six signal-controlled junctions.
decision variables handled were the cycle time and green Experiments showed that the resulting values for degree of
time. The performance of the improved ACO was compared to saturation indicated that none of the links were oversaturated,
the Mutually Consistent (MC) solution for optimising signal and the network was uncongested.
timings. Both algorithms were tested on an artificial network The work in [110] proposes a discrete harmony
layout composed of 20 paths and 6 signalized junctions. The search (DHS) algorithm to minimize the network-wise total
ACO algorithm was found to be highly successful in terms delay time in a finite time horizon. Moreover, an ensemble of
of the signal timings and the reached degrees of saturation. local search operators was proposed to enhance the algorithm’s
Although the MC solution gives the similar results in terms performance. A single harmony (i.e. solution) represents the
of the value of objective function, it resulted in greater cycle stages of all intersections in a one-time window. Different local
times compared to the bi-level solution. search operators were introduced to randomly change some

Authorized licensed use limited to: SASI INSTITUTE OF TECHNOLOGY & ENGINEERING. Downloaded on October 23,2024 at 05:35:24 UTC from IEEE Xplore. Restrictions apply.
 SHAIKH et al.: REVIEW ON SWARM INTELLIGENCE AND EVOLUTIONARY ALGORITHMS 59

stage numbers in some harmonies. Teste cases with different the same solution representation and objective function in
window sizes were generated based on real-life traffic data. [110], [111]. Experimental results revealed the superiority
DHS was compared to the fixed cycle traffic light control of applied metaheuristics against the existing traffic control
system (FCS) in Singapore. Compared to FCS, DHS has system, with an improvement ranging from over 26% and
provided an improvement of 17.9% and 30.3% in the average 28% in terms of minimum and maximum total delay times.
delay time across all test cases for window sizes of 30s and Moreover, the WCA was able to obtain better statistical
60s, respectively. optimization results for with the HS and Jaya except for
DHS was also adopted in [111] to minimize the total smaller cases. Furthermore, WCA provided less average CPU
network cost for both pedestrians and vehicles. Another goal time.
was to overcome the high computational complexity involved The works in [110], [111] has been further extended in [119]
in real-time traffic light scheduling. When compared to optimal to optimize both the network-wise total delay of all vehicles
solutions obtained by solving the proposed MILP model, DHS and the total delay time of all pedestrians. The authors
running times were in seconds while providing a solution, implemented both DHS and Discrete ABC (DABC) with
which always has less than a 25% gap from the optimal solu- local search operators to solve the problem. Experiments were
tion. This indicated that DHS can provide a significant increase conducted on 19 test cases using real data from the Jurong
of efficiency meanwhile ensuring an acceptable degree of area in Singapore. Solution representation remained the same
performance degradation. with the difference of optimizing two objectives. The inverted
In [112], the authors developed a CTM-based traffic signal generational distance (IGD) provided by both algorithms were
scheduling strategy with consideration of drivers’ unhappiness. far less than those of the currently used traffic light control
When only the drivers’ unhappiness (i.e. drivers’ waiting strategy with DABC having the best performance. Moreover,
time), DHS outperformed the fixed-time strategy with increas- both DHS and DABC had better performance than NSGA-II
ing vehicle incoming flow. Afterwards, a bi-objective opti- with and without the local search operator.
mization problem formulation was adopted to minimize the In [120], the authors developed a model to describe a het-
drivers’ waiting time (i.e. drivers’ unhappiness) and the total erogeneous traffic network with signalized and non-signalized
network delay (i.e. traffic performance). Comparison were run intersections. The objective was to minimize the network-
between NSGA-II and Non-dominated Sorting HS (NSHS) wise time delay of all vehicles. Five metaheuristic algorithms
on networks of different sizes and different vehicles volume. were compared (with and without ensemble of local search
It was shown that NSGA-II outperformed NSHS on all cases. operators): GA, ABC, DHS, Jaya, and WCA. All algorithm
used the same solution representation in [110], [111]. Algo-
VI. C OMPARATIVE S TUDIES rithms were evaluates using 18 cases, based on real data,
This section surveys previous works in the literature that with different scales. The algorithm with the best performance
aimed at comparing different EA/SI techniques when applied was ABC with local search followed by DHS, GA, Jaya, and
to the TSC problem. WCA. In addition, ABC produced an average improvement
The work in [113] compared the performance of three of 26.3% when compared the currently used traffic signals
different GAs and PSO when applied to a 13-traffic-lights control strategy in Singapore, which is which is a fixed cycle
network. The aim was to optimize the traffic light signal tim- control strategy with dynamic updating using a geomagnetic
ings to minimize the total travel time. Algorithms’ parameters sensor.
were first tuned using experiments on a simpler benchmark The work in [121] proposed a bi-level model to automati-
function. Experiments showed that parameter tuning plays a cally estimate the time-of-day intervals (at the upper-level) and
very important role in the performance as some parameters the traffic control problems over these intervals (at the lower-
that were efficient in one experiment weren’t providing good level). The decision variables where the control strategies (the
results in other experiments. cycle length and effective green times) and the period where
In [114], the authors conducted a comparative analytical the strategy is applied. the objective was to minimize the total
study of PSO, DE, and CMA-ES (Evolutionary Strategy delay. The authors tested PSO, GA, and SA with the Nelder-
with Covariance Matrix Adaptation) [115] when applied to Mead method for local search. It was shown that the proposed
traffic optimization. Experiments were run using an artificial hybridization speeded up the original algorithms and that GA
8-junction network with every intersection having two phases. with NM has produced the best results.
The objective was formulated using weighted average to mini- The study in [122] aimed to develop metaheuristic-based
mize the number of stops and maximize the number of vehicles methods for intelligent traffic control at isolated signalized
reaching their destination. The algorithms optimized the phase intersections, in the city of Dhahran, Saudi Arabia. Both GA
lengths of the intersection with and without the offset. Results and DE were used to optimize the signal timings plan to reduce
reveled that all algorithms provided better results than the map the average vehicle delay. Decision variables manipulated were
editor in the SUMO package with CMA-ES providing better Green splits in each phase. Results illustrated that both GA and
results in both scenarios. DE produced systematic signal timings plans and significantly
The work in [116] compared the performance of Jaya reduced travel time delay ranging from 15% to 35% compared
Algorithm [117], Water Cycle Algorithm (WCA) [118], and to existing conditions. Moreover, DE converged much faster
HS when applied to 11 test cases generated based on real- than GA although it was outperformed by GA in terms of
traffic data in Singapore. Algorithms were implemented using solution quality.

Authorized licensed use limited to: SASI INSTITUTE OF TECHNOLOGY & ENGINEERING. Downloaded on October 23,2024 at 05:35:24 UTC from IEEE Xplore. Restrictions apply.
 60 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 23, NO. 1, JANUARY 2022

VII. C ONCLUSION AND F UTURE D IRECTIONS a proposed algorithm is properly evaluated and would
The current paper provided an overall literature review of provide the research community with a baseline against
the state-of-the-art research on addressing traffic signal control which algorithms could be fairly evaluated.
(TSC) problem using evolutionary algorithms, swarm intelli- 6) Simple and efficient EA and SI algorithms are key to
gence algorithms, and other population-based meta-heuristics. solving TSC effectively. Designing ensembles of various
Surveyed works were compared based on the selected decision strategies [125] to improve the algorithms’ performance
variables, the optimized objective(s), the problem formulation is a promising direction to be investigated.
(i.e. single objective, multi-objective, or bi-level) and the 7) Using cooperative optimization frameworks like Coop-
solution encoding. erative Co-evolution (CC) [126] enable the decomposi-
Extracted data revealed an increasing interest, over the last tion of the problem (i.e. traffic network) into separate
two decades, in applying EA/SI to TSC. Based on the surveyed components (i.e. sub-networks) that are co-optimized
work, a number of promising future research directions are cooperatively. Such a framework can allow for reaching
identified: better solutions in less time specially for large-scale
1) Almost all surveyed works do not consider some real- optimization problems (i.e. traffic networks with large
world constraints and dynamic events (e.g., accidents, number of intersections).
emergency vehicles, evacuation events, weather condi- 8) One important direction towards enabling EA/SI imple-
tions …etc.). These must be considered if TSC is to be mentations to be applied in real-life is to reduce their
solved in real-life environments. By considering such running times. Although some works adopted paral-
real-world scenarios, theoretical research using EA/SI lelized implementations [61], [75], [95], this aspect
could be applied to specific areas or specific urban traffic is still lacking. Even if computational nodes are
networks. not readily available or accessible, Graphic Process-
2) Considering objective functions based on energy con- ing Units (GPUs) offer a promising implementation
sumption is a hot topic in transportation in general. platform [127].
In [123], the authors highlight that metaheuristics com- 9) As the current work is only focused on manipulating
bined with machine learning methods can provide the traffic signal schedules/timings using EA/SI, more stud-
required flexibility to deal with uncertainties and the ies could extend on this to include EA/SI applications to
dynamic nature of complex transportation problems other areas in transportation. Furthermore, comparisons
involving energy optimization. A number of surveyed could be conducted against available exact methods.
works addressed the issue of emissions [60], [66], [77], R EFERENCES
[86], [104], [108], however, this only constitutes around
[1] 68% of the World Population Projected to Live in Urban Areas
13% of the covered literature. by 2050, Says UN|UN DESA Department of Economic and Social
3) A vast majority of the surveyed works, around 72%, Affairs. Accessed: May 16, 2018. [Online]. Available: https://www.un.
tackle TSC as a single objective problem (using weights org/development/desa/en/news/population/2018-revision-of-world-
urbanization-prospects.html
for multiple objectives). Multiple objectives, or even [2] ACEA Report Vehicles in Use Europe 2017, Eur. Automobile Manuf.
many objectives, have to be optimized to properly track Assoc., Brussels, Belgium, 2017.
the pareto front and satisfy different performance indi- [3] Passenger Cars in Use Worldwide 2005-2015|Statistic. Accessed:
Aug. 5, 2020. [Online]. Available: https://www.statista.com/statistics/
cators. Optimization strategies for multi-objective and 274365/passenger-cars-in-use-worldwide/
many-objective problems remain a challenging issue and [4] INRIX Global Traffic Scorecard. Accessed: Aug. 5, 2020. [Online].
is worthy of further studies for TSC. Available: http://inrix.com/scorecard/
[5] M. S. Shirazi and B. T. Morris, “Looking at intersections: A survey of
4) The collaboration between TSC, drivers, and pedestrians intersection monitoring, behavior and safety analysis of recent studies,”
is an important link in intelligent transportation sys- IEEE Trans. Intell. Transp. Syst., vol. 18, no. 1, pp. 4–24, Jan. 2017.
tems. Communications between TSC and the driving [6] S. C. Wong, W. T. Wong, C. M. Leung, and C. O. Tong, “Group-based
optimization of a time-dependent TRANSYT traffic model for area
system (human drivers) [111], pedestrians [111], [124], traffic control,” Transp. Res. B, Methodol., vol. 36, no. 4, pp. 291–312,
infrastructures and/or autonomous vehicles still pose May 2002.
huge challenges. [7] H. J. Ruskin and R. Wang, “Modeling traffic flow at an urban
unsignalized intersection,” in Proc. Int. Conf. Comput. Sci. Berlin,
5) One clear aspect that is lacking in the domain of TSC is Germany: Springer, 2002, pp. 381–390.
the availability of common libraries for benchmarking [8] E. Chevallier and L. Leclercq, “A macroscopic theory for unsignal-
ized intersections,” Transp. Res. B, Methodol., vol. 41, no. 10,
and comparing different algorithms. A vast majority of pp. 1139–1150, Dec. 2007.
publications apply proposed algorithms on their own [9] W. Brilon and N. Wu, “Capacity at unsignalized intersections derived
generated artificial networks and/or networks from the by conflict technique,” Transp. Res. Rec., J. Transp. Res. Board,
vol. 1776, no. 1, pp. 82–90, Jan. 2001.
authors’ cities/countries without comparing with previ- [10] N. Wu, “A universal procedure for capacity determination at unsignal-
ous work. This poses a huge challenge in identifying ized (priority-controlled) intersections,” Transp. Res. B, Methodol.,
the state-of-the-art in TSC. The current literature review vol. 35, no. 6, pp. 593–623, Jul. 2001.
was presented in chronological order and rarely it is [11] N. Wu, “A new model for estimating critical gap and its distribution at
unsignalized intersections based on the equilibrium of probabilities,”
found that authors provide comparisons with existing in Proc. 5th Int. Symp. Highway Capacity Qual. Service, Jul. 2006,
works. Such libraries should contain networks of differ- pp. 1–10.
ent sizes and under different conditions (undersaturated, [12] M. E. Foulaadvand and S. Belbasi, “Vehicular traffic flow at a
nonsignalized intersection,” J. Phys. A, Math. Theor., vol. 40, no. 29,
saturated, and oversaturated). This would ensure that p. 8289. 2007.

Authorized licensed use limited to: SASI INSTITUTE OF TECHNOLOGY & ENGINEERING. Downloaded on October 23,2024 at 05:35:24 UTC from IEEE Xplore. Restrictions apply.
 SHAIKH et al.: REVIEW ON SWARM INTELLIGENCE AND EVOLUTIONARY ALGORITHMS 61

[13] Q.-S. Wu, X.-B. Li, M.-B. Hu, and R. Jiang, “Study of traffic flow [41] K. S. Lee and Z. W. Geem, “A new meta-heuristic algorithm for
at an unsignalized T-shaped intersection by cellular automata model,” continuous engineering optimization: Harmony search theory and
Eur. Phys. J. B, vol. 48, no. 2, pp. 265–269, Nov. 2005. practice,” Comput. Methods Appl. Mech. Eng., vol. 194, nos. 36–38,
[14] C. Chen, F. Zhu, Y. Ai, and H. Yunfeng, “A survey of urban traffic pp. 3902–3933, Sep. 2005.
signal control for agent recommendation system,” in Proc. IEEE Conf. [42] C. Blum and A. Roli, “Metaheuristics in combinatorial optimization:
Intell. Transp., Sep. 2012, pp. 327–333. Overview and conceptual comparison,” ACM Comput. Surveys, vol. 35,
[15] F. Webster, Traffic Signal Settings Road Research Technical Paper. no. 3, pp. 268–308, Sep. 2003.
London, U.K.: H. M. Stationery Off, 1958. [43] A. E. Eiben and E. James Smith, Introduction to Evolutionary Com-
[16] D. C. John Little, D. Mark Kelson, and H. Nathan Gartner, puting. Cham, Switzerland: Springer, 2003.
“MAXBAND: A versatile program for setting signals on arteries and [44] P. A. Engelbrecht, Fundamentasls of Computational Swarm Intelli-
triangular networks,” Massachusetts Inst. Technol., Cambridge, MA, gence. Hoboken, NJ, USA: Wiley, 2005.
USA, Tech. Rep. 1185-81, 1981. [45] A. Slowik and H. Kwasnicka, “Nature inspired methods and their
[17] D. I. Robertson, “TRANSYT method for area traffic control,” Traffic industry Applications—Swarm intelligence algorithms,” IEEE Trans.
Eng. Control, vol. 11, no. 6, pp. 276–281, 1969. Ind. Informat., vol. 14, no. 3, pp. 1004–1015, Mar. 2018.
[18] P. B. Hunt, D. L. Robertson, and R. D. Bretherton, “The SCOOT on- [46] L. Brezoänik, I. Fister, and V. Podgorelec, “Swarm intelligence algo-
line traffic signal optimization technique,” Traffic Eng. Control, vol. 23, rithms for feature selection: A review,” Appl. Sci., vol. 8, no. 9, p. 1521,
no. 4, pp. 190–192, 1982. Sep. 2018.
[19] N. H. Gartner, “OPAC: A demand-responsive strategy for traffic signal [47] C. Kolias, G. Kambourakis, and M. Maragoudakis, “Swarm intelligence
control,” in Proc. 62nd Annu. Meeting Transp. Res. Board, 1983, in intrusion detection: A survey,” Comput. Secur., vol. 30, no. 8,
pp. 75–81. pp. 625–642, Nov. 2011.
[20] J. L. Farges, J. J. Henry, and J. Tufal, “The PRODYN real-time traffic [48] M. Mavrovouniotis, C. Li, and S. Yang, “A survey of swarm intelli-
algorithm,” Proc. 4th IFAC/IFIP/IFORS Conf., Control Transp. Syst., gence for dynamic optimization: Algorithms and applications,” Swarm
1984, pp. 305–310. Evol. Comput., vol. 33, pp. 1–17, Apr. 2017.
[21] F. Boillot, J. M. Blosseville, J. B. Lesort, V. Motyka, M. Papageorgiou, [49] B. Y. Qu, Y. S. Zhu, Y. C. Jiao, M. Y. Wu, P. N. Suganthan, and
and S. Sellam, “Optimal signal control of urban traffic networks,” in J. J. Liang, “A survey on multi-objective evolutionary algorithms for
Proc. Road Traffic Monit., 1992, p. 35. the solution of the environmental/economic dispatch problems,” Swarm
[22] S. Sen and K. L. Head, “Controlled optimization of phases at an Evol. Comput., vol. 38, pp. 1–11, Feb. 2018.
intersection,” Transp. Sci., vol. 31, no. 1, pp. 5–17, Feb. 1997. [50] M. A. Dulebenets, “A comprehensive evaluation of weak and strong
[23] M. J. Lighthill and G. B. Whitham, “On kinematic waves. II. A theory mutation mechanisms in evolutionary algorithms for truck scheduling
of traffic flow on long crowded roads,” Proc. Roy. Soc. London, Ser. at cross-docking terminals,” IEEE Access, vol. 6, pp. 65635–65650,
A, Math. Phys. Sci., vol. 229, no. 1178, pp. 317–345, 1955. 2018.
[24] P. I. Richards, “Shock waves on the highway,” Oper. Res., vol. 4, no. 1, [51] M. A. Dulebenets, “A delayed start parallel evolutionary algorithm for
pp. 42–51, Feb. 1956. just-in-time truck scheduling at a cross-docking facility,” Int. J. Prod.
[25] C. F. Daganzo, “The cell transmission model: A dynamic representation Econ., vol. 212, pp. 236–258, Jun. 2019.
of highway traffic consistent with the hydrodynamic theory,” Transp. [52] Z. Liu, “A survey of intelligence methods in urban traffic signal
Res. B, Methodol., vol. 28, no. 4, pp. 269–287, Aug. 1994. control,” Int. J. Comput. Sci. Netw. Secur., vol. 7, no. 7, pp. 105–112,
2007.
[26] C. F. Daganzo, “A finite difference approximation of the kinematic
wave model of traffic flow,” Transp. Res. B, Methodol., vol. 29, no. 4, [53] D. Zhao, Y. Dai, and Z. Zhang, “Computational intelligence in urban
pp. 261–276, Aug. 1995. traffic signal control: A survey,” IEEE Trans. Syst., Man, Cybern.
C, Appl. Rev., vol. 42, no. 4, pp. 485–494, Jul. 2012.
[27] H. K. Lo, “A novel traffic signal control formulation,” Transp. Res.
A, Policy Pract., vol. 33, no. 6, pp. 433–448, Aug. 1999. [54] S. Araghi, A. Khosravi, and D. Creighton, “A review on computational
intelligence methods for controlling traffic signal timing,” Expert Syst.
[28] K. Hong Lo, “A cell-based traffic control formulation: Strategies
Appl., vol. 42, no. 3, pp. 1538–1550, Feb. 2015.
and benefits of dynamic timing plans,” Transp. Sci., vol. 35, no. 2,
[55] P. Mannion, J. Duggan, and E. Howley, “An experimental review of
pp. 148–164, 2001.
reinforcement learning algorithms for adaptive traffic signal control,”
[29] L. Qi, M. Zhou, and W. Luan, “Modeling and control of urban road
in Autonomic Road Transport Support Systems. Autonomic Systems,
intersections with incidents via timed Petri nets,” in Proc. IEEE 12th
T. McCluskey, A. Kotsialos, J. Müller, F. Klägl, O. Rana, and
Int. Conf. Netw., Sens. Control, Apr. 2015, pp. 185–190.
R. Schumann, Eds. Basel, Switzerland: Birkhäuser, 2016, pp. 47–66.
[30] M. H. Yar, V. Rahmati, and H. R. D. Oskouei, “A survey on evolu- [56] K.-L.-A. Yau, J. Qadir, H. L. Khoo, M. H. Ling, and P. Komisarczuk,
tionary computation: Methods and their applications in engineering,” “A survey on reinforcement learning models and algorithms for traffic
Modern Appl. Sci., vol. 10, no. 11, p. 131, Aug. 2016. signal control,” ACM Comput. Surv., vol. 50, no. 3, pp. 1–38, Oct. 2017.
[31] A. E. Eiben and J. E. Smith, Introduction to Evolutionary Computing, [57] H. Wei, G. Zheng, V. Gayah, and Z. Li, “A survey on traffic sig-
vol. 53. Berlin, Germany: Springer, 2003. nal control methods,” 2019, arXiv:1904.08117. [Online]. Available:
[32] J. Zhang et al., “Evolutionary computation meets machine learning: http://arxiv.org/abs/1904.08117
A survey,” IEEE Comput. Intell. Mag., vol. 6, no. 4, pp. 68–75, [58] Y. Wang, X. Yang, H. Liang, and Y. Liu, “A review of the self-adaptive
Nov. 2011. traffic signal control system based on future traffic environment,”
[33] Darwinism. Wikipedia. Accessed: Oct. 31, 2018. [Online]. Available: J. Adv. Transp., vol. 2018, pp. 1–12, Jun. 2018.
https://en.wikipedia.org/wiki/Darwinism [59] S. Srivastava and S. K. Sahana, “A survey on traffic optimization
[34] J. Kennedy, R. C. Eberhart, “Particle swarm optimization,” in Proc. problem using biologically inspired techniques,” Natural Comput.,
IEEE Int. Conf. Neural Netw., vol. 4, May 1995, pp. 1942–1948. vol. 10, pp. 1–15, Jan. 2019.
[35] R. C. Eberhart and J. Kennedy, “A new optimizer using particle [60] B. Zhang, L. Shang, and D. Chen, “Traffic intersection signal-planning
swarm theory,” in Proc. 6th Int. Symp. Micro Mach. Hum. Sci., 1995, multi-object optimization based on genetic algorithm,” in Proc. 2nd Int.
pp. 39–43. Workshop Intell. Syst. Appl., May 2010, pp. 1–4.
[36] M. Dorigo, “Optimization, learning and natural algorithms,” [61] J. Javier Sánchez-Medina, J. Manuel Calán-Moreno, and E. Rubio-
Ph.D. dissertation, Dept. di Elettronica, Politecnico di Milano, Royo, “Traffic signal optimization in ‘La Almozara’ district in
Milano, Italy, 1992. Saragossa under congestion conditions, using genetic algorithms, traffic
[37] D. Karaboga, “An idea based on honey bee swarm for numerical opti- microsimulation, and cluster computing,” IEEE Trans. Intell. Transp.
mization,” Eng. Faculty, Comput. Eng. Dept., Erciyes Univ., Kayseri, Syst., vol. 11, no. 1, pp. 132–141, May 2010.
Turkey, Tech. Rep. TR06, 2005. [62] Y. K. Chin, K. C. Yong, N. Bolong, S. S. Yang, and K. T. K. Teo,
[38] X. S. Yang, S. Deb, “Cuckoo search via Lévy flights,” in “Multiple intersections traffic signal timing optimization with genetic
Proc. World Congr. Nature Biologically Inspired Comput., 2009, algorithm,” in Proc. IEEE Int. Conf. Control Syst., Comput. Eng.,
pp. 210–214. Nov. 2011, pp. 454–459.
[39] X. S. Yang, “A new metaheuristic bat-inspired algorithm,” in Proc. [63] Z. Shen, K. Wang, and F.-Y. Wang, “GPU based non-dominated
Nature Inspired Cooperat. Strategies Optim., 2010, pp. 65–74. sorting genetic algorithm-II for multi-objective traffic light signaling
[40] Y. Shi, “An optimization algorithm based on brainstorming process,” optimization with agent based modeling,” in Proc. 16th Int. IEEE Conf.
Int. J. Swarm Intell. Res., vol. 2, no. 4, pp. 35–62, Oct. 2011. Intell. Transp. Syst. (ITSC), Oct. 2013, pp. 1840–1845.

Authorized licensed use limited to: SASI INSTITUTE OF TECHNOLOGY & ENGINEERING. Downloaded on October 23,2024 at 05:35:24 UTC from IEEE Xplore. Restrictions apply.
 62 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 23, NO. 1, JANUARY 2022

[64] K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, “A fast and [87] E. Ricalde and W. Banzhaf, “A genetic programming approach for the
elitist multiobjective genetic algorithm: NSGA-II,” IEEE Trans. Evol. traffic signal control problem with epigenetic modifications,” in Proc.
Comput., vol. 6, no. 2, pp. 182–197, Apr. 2002. Eur. Conf. Genetic Program., 2016, pp. 133–148.
[65] Y. Li, L. Yu, S. Tao, and K. Chen, “Multi-objective optimization of [88] E. Ricalde and W. Banzhaf, “Evolving adaptive traffic signal controllers
traffic signal timing for oversaturated intersection,” Math. Problems for a real scenario using genetic programming with an epigenetic mech-
Eng., vol. 2013, Dec. 2013, Art. no. 182643. anism,” in Proc. 16th IEEE Int. Conf. Mach. Learn. Appl. (ICMLA),
[66] R. Qian, Z. Lun, Y. Wenchen, and Z. Meng, “A traffic emission Dec. 2017, pp. 897–902.
saving signal timing model for urban isolated intersections,” Proc. [89] C. Dong, S. Huang, and X. Liu, “Urban area traffic signal timing
Socia Behav. Sci., vol. 96, pp. 2404–2413, Oct. 2013. optimization based on sa-PSO,” in Proc. Int. Conf. Artif. Intell. Comput.
[67] H. Dezani, N. Marranghello, and F. Damiani, “Genetic algorithm-based Intell., Oct. 2010, pp. 80–84.
traffic lights timing optimization and routes definition using Petri net [90] J. García-Nieto, E. Alba, and A. Carolina Olivera, “Swarm intelligence
model of urban traffic flow,” in Proc. 19th World Congr. Int. Fed. for traffic light scheduling: Application to real urban areas,” Eng. Appl.
Autom. Control, 2014, pp. 11326–11331. Artif. Intell., vol. 25, no. 2, pp. 274–283, Mar. 2012.
[68] H. Dezani, R. D. S. Bassi, N. Marranghello, L. Gomes, F. Damiani, [91] J. Garcia-Nieto, A. C. Olivera, and E. Alba, “Optimal cycle program
and I. Nunes da Silva, “Optimizing urban traffic flow using genetic of traffic lights with particle swarm optimization,” IEEE Trans. Evol.
algorithm with Petri net analysis as fitness function,” Neurocomputing, Comput., vol. 17, no. 6, pp. 823–839, Dec. 2013.
vol. 124, pp. 162–167, Jan. 2014. [92] Z. Yong and W. Peng, “Optimal control for region of the city traffic
[69] H.-Y. Tung, W.-C. Ma, and T.-L. Yu, “Novel traffic signal timing signal based on selective particle swarm optimization algorithm,” in
adjustment strategy based on genetic algorithm,” in Proc. IEEE Congr. Proc. Int. Conf. Intell. Transp., Big Data Smart City, Dec. 2015,
Evol. Comput. (CEC), Jul. 2014, pp. 2353–2360. pp. 544–550.
[70] M. K. Tan, H. S. E. Chuo, R. K. Y. Chin, K. B. Yeo, and [93] I. Geda Pasek Suta Wijaya, K. Uchimura, and G. Koutaki, “Traffic light
K. T. K. Teo, “Optimization of urban traffic network signalization using signal parameters optimization using particle swarm optimization, case
genetic algorithm,” in Proc. IEEE Conf. Open Syst. (ICOS), Oct. 2016, study of ooe toroku road network optimization,” in Proc. Int. Seminar
pp. 87–92. Intell. Technol. Appl., 2015, pp. 11–16.
[71] H. Jung, S. Choi, B. B. Park, H. Lee, and S. H. Son, “Bi-level opti- [94] I. G. P. S. Wijaya, “The improvement me-ga based traffic signal control
mization for eco-traffic signal system,” in Proc. Int. Conf. Connected optimization using new fitness model,” GSTF J. Comput., vol. 2, no. 2,
Vehicles Expo (ICCVE), Sep. 2016. pp. 64–69, 2012.
[72] V. Oduguwa and R. Roy, “Bi-level optimisation using genetic algo- [95] K. Han, Y. Sun, H. Liu, T. L. Friesz, and T. Yao, “A bi-level model of
rithm,” in Proc. IEEE Int. Conf. Artif. Intell. Syst. (ICAIS), 2002, dynamic traffic signal control with continuum approximation,” Transp.
pp. 322–327. Res. C, Emerg. Technol., vol. 55, pp. 409–431, Jun. 2015.
[73] P. Jiao, H. Wang, and T. Sun, “Real-time arterial coordination [96] S. Dabiri and M. Abbas, “Arterial traffic signal optimization using parti-
control based on dynamic intersection turning fractions estimation cle swarm optimization in an integrated VISSIM-MATLAB simulation
using genetic algorithm,” Math. Problems Eng., vol. 2014, Jul. 2014, environment,” in Proc. IEEE 19th Int. Conf. Intell. Transp. Syst. (ITSC),
Art. no. 430497. Nov. 2016, pp. 766–771.
[74] X. Li, Z. Zhao, L. Liu, Y. Liu, and P. Li, “An optimization model [97] B. Wu and D. Wang, “Traffic signal networks control optimize with
of multi-intersection signal control for trunk road under collaborative PSO algorithm,” in Proc. 12th Int. Conf. Natural Comput., Fuzzy Syst.
information,” J. Control Sci. Eng., vol. 2017, pp. 1–11, Apr. 2017. Knowl. Discovery (ICNC-FSKD), Aug. 2016, pp. 230–234.
[75] Z. Li, M. Shahidehpour, S. Bahramirad, and A. Khodaei, “Optimizing [98] G. Tong, F. Cui, and C. Fan, “Genetic algorithm and its application in
traffic signal settings in smart cities,” IEEE Trans. Smart Grid, vol. 8, the realtime traffic signal optimization control,” in Proc. 6th Int. Conf.
no. 5, pp. 2382–2393, Sep. 2017. ITS Telecommun., Jun. 2006, pp. 86–89.
[76] N. R. Sabar, L. M. Kieu, E. Chung, T. Tsubota, and P. E. Maciel [99] W. Hu, H. Wang, L. Yan, and B. Du, “A swarm intelligent method for
de Almeida, “A memetic algorithm for real world multi-intersection traffic light scheduling: Application to real urban traffic networks,” Int.
traffic signal optimisation problems,” Eng. Appl. Artif. Intell., vol. 63, J. Speech Technol., vol. 44, no. 1, pp. 208–231, Jan. 2016.
pp. 45–53, Aug. 2017. [100] P. Jiao, R. Li, and Z. Li, “Pareto front-based multi-objective real-
[77] W. Kou, X. Chen, L. Yu, and H. Gong, “Multiobjective optimization time traffic signal control model for intersections using particle swarm
model of intersection signal timing considering emissions based on optimization algorithm,” Adv. Mech. Eng., vol. 8, no. 8, pp. 1–15, 2016.
field data: A case study of beijing,” J. Air Waste Manage. Assoc., [101] H. S. E. Chuo, M. K. Tan, A. C. H. Chong, R. K. Y. Chin, and
vol. 68, no. 8, pp. 836–848, Aug. 2018. K. T. K. Teo, “Evolvable traffic signal control for intersection conges-
[78] X. Li and J.-Q. Sun, “Signal multiobjective optimization for urban tion alleviation with enhanced particle swarm optimisation,” in Proc.
traffic network,” IEEE Trans. Intell. Transp. Syst., vol. 19, no. 11, IEEE 2nd Int. Conf. Autom. Control Intell. Syst. (ICACIS), Oct. 2017,
pp. 3529–3537, Nov. 2018. pp. 92–97.
[79] X. Li and J.-Q. Sun, “Multi-objective optimal predictive control of [102] M. A. Chentoufi and R. Ellaia, “A hybrid particle swarm optimization
signals in urban traffic network,” J. Intell. Transp. Syst., vol. 23, no. 4, and tabu search algorithm for adaptive traffic signal timing optimiza-
pp. 370–388, Jul. 2019. tion,” in Proc. IEEE Int. Conf. Technol. Manage., Oper. Decisions
[80] C. Ma and P. Liu, “Intersection signal timing optimization considering (ICTMOD), Nov. 2018, pp. 25–30.
the travel safety of the elderly,” Adv. Mech. Eng., vol. 11, no. 12, [103] X. Li and D. Li, “Dynamic self-adaptive chaotic particle swarm
pp. 1–8, 2019. optimization algorithm in optimal control of public transit priority
[81] H. Ceylan, “Optimal design of signal controlled road networks using traffic signal,” in Proc. 10th Int. Conf. Intell. Hum.-Mach. Syst. Cybern.
differential evolution optimization algorithm,” Math. Problems Eng., (IHMSC), Aug. 2018, pp. 208–213.
vol. 2013, Aug. 2013, Art. no. 696374. [104] H. Jia, Y. Lin, Q. Luo, Y. Li, and H. Miao, “Multi-objective optimiza-
[82] R. E. Allsop and J. A. Charlesworth, “Traffic in a signalcontrolled road tion of urban road intersection signal timing based on particle swarm
network: An example of different signal timings including different optimization algorithm,” Adv. Mech. Eng., vol. 11, no. 4, pp. 1–9, 2019.
routings,” Traffic Eng. Control, vol. 18, no. 5, pp. 262–265, 1977. [105] E. Zechman, L. Quadrifoglio, and R. Putha, “Ant colony optimiza-
[83] H. Ceylan and M. G. H. Bell, “Traffic signal timing optimisation based tion algorithm for signal coordination of oversaturated traffic net-
on genetic algorithm approach, including drivers’ routing,” Transp. Res. works,” Texas Transp. Inst., Texas Univ., College Station, TX, USA,
B, Methodol., vol. 38, no. 4, pp. 329–342, 2004. Tech. Rep. SWUTC/10/169113-1, 2010.
[84] H. Ceylan and H. Ceylan, “A hybrid harmony search and TRANSYT [106] O. Baskan and S. Haldenbilen, “Ant colony optimization approach for
hill climbing algorithm for signalized stochastic equilibrium trans- optimizing traffic signal timings,” in Proc. Ant Colony Optimization-
portation networks,” Transp. Res. C, Methodol., vol. 25, pp. 152–167, Methods Appl., 2011, pp. 205–220.
Oct. 2012. [107] D. Renfrew and X.-H. Yu, “Traffic signal optimization using ant colony
[85] Z. Cakici and Y. S. Murat, “A differential evolution algorithm-based algorithm,” in Proc. Int. Joint Conf. Neural Netw. (IJCNN), Jun. 2012,
traffic control model for signalized intersections,” Adv. Civil Eng., pp. 1–7.
vol. 2019, Oct. 2019, Art. no. 7360939. [108] R. Peñabaena-Niebles, V. Cantillo, J. L. Moura, and A. Ibeas, “Design
[86] O. Baskan, “A multiobjective bilevel programming model for environ- and evaluation of a mathematical optimization model for traffic signal
mentally friendly traffic signal timings,” Adv. Civil Eng., vol. 2019, plan transition based on social cost function,” J. Adv. Transp., vol. 2017,
pp. 1–13, Oct. 2019. pp. 1–12, Oct. 2017.

Authorized licensed use limited to: SASI INSTITUTE OF TECHNOLOGY & ENGINEERING. Downloaded on October 23,2024 at 05:35:24 UTC from IEEE Xplore. Restrictions apply.
 SHAIKH et al.: REVIEW ON SWARM INTELLIGENCE AND EVOLUTIONARY ALGORITHMS 63

[109] M. Dell’Orco, O. Baskan, and M. Marinelli, “A harmony search algo- Palwasha W. Shaikh (Member, IEEE) received the B.Eng. degree (summa
rithm approach for optimizing traffic signal timings,” Traffic Transp., cum laude) in computer engineering from the American University of Kuwait
vol. 25, no. 4, pp. 349–358, Jul. 2013. (AUK), Salmiya, Kuwait, in 2019. She is currently pursuing the M.Sc. degree
[110] K. Gao, Y. Zhang, A. Sadollah, and R. Su, “Optimizing urban traffic in electrical and computer engineering with the University of Ottawa (UO),
light scheduling problem using harmony search with ensemble of local Ottawa, ON, Canada. From 2015 to 2019, she was a Teaching Assistant with
search,” Appl. Soft Comput., vol. 48, pp. 359–372, Nov. 2016. the Tutoring Center (TRC), AUK. From 2016 to 2019, she was a Research
[111] Y. Zhang, K. Gao, Y. Zhang, and R. Su, “Traffic light scheduling for Assistant with AUK with the Department of Engineering. She is currently
pedestrian-vehicle mixed-flow networks,” IEEE Trans. Intell. Transp. a Research Assistant with UO under Dr. Hussein T. Mouftah’s supervision.
Syst., vol. 20, no. 4, pp. 1468–1483, Apr. 2019. She has researched on embedded systems, traffic signal control problems,
[112] Y. Zhang, Y. Zhang, and R. Su, “Optimization of vehicle delay light-fidelity, IoT and connected vehicles in the past. Her current research
and drivers’ unhappiness at a signalized network: A multi-objective interests include smart connected and autonomous electric vehicles, deep
approach,” in Proc. IEEE 15th Int. Conf. Control Autom. (ICCA), learning, wireless sensor networks, smart grids, and dynamic wireless charging
Jul. 2019, pp. 622–627. infrastructure and schemes for connected and autonomous electric vehicles.
[113] R. K. Abushehab, B. K. Abdalhaq, and B. Sartawi, “Genetic vs. Particle
swarm optimization techniques for traffic light signals timing,” in Proc.
6th Int. Conf. Comput. Sci. Inf. Technol. (CSIT), Mar. 2014, pp. 27–35.
[114] K. Malecki and S. Iwan, “Comparative analysis of selected algorithms Mohammed El-Abd (Senior Member, IEEE) received the B.Eng. and M.Sc.
in the process of optimization of traffic lights,” in Proc. Asian Conf. degrees from the ECE Department, Ain Shams University, Egypt, in 1998
Intell. Inf. Databse Syst., 2017, pp. 497–506. and 2003, respectively, and the Ph.D. degree from the ECE Department,
[115] N. Hansen, “The CMA evolution strategy: A tutorial,” 2016, University of Waterloo (UW), Canada, in 2008. He is currently an Associate
arXiv:1604.00772. [Online]. Available: http://arxiv.org/abs/1604.00772 Professor of computer engineering with the Department of Engineering, The
[116] K. Gao, Y. Zhang, A. Sadollah, A. Lentzakis, and R. Su, “Jaya, American University of Kuwait (AUK). He has published more than 10 journal
harmony search and water cycle algorithms for solving large-scale articles and over 60 conference publications, abstracts, and book chapters.
real-life urban traffic light scheduling problem,” Swarm Evol. Comput., He is an Associate Editor of the Swarm and Evolutionary Computation
vol. 37, pp. 58–72, Dec. 2017. (SWEVO) journal by Elsevier since 2016 and served as a Guest Editor for
[117] R. V. Rao, “Jaya: A simple and new optimization algorithm for the IEEE T RANSACTIONS ON E DUCATION. He serves as a reviewer for many
solving constrained and unconstrained optimization problems,” Int. prestigious journals including the IEEE T RANSACTIONS ON E VOLUTIONARY
J. Ind. Eng. Comput., vol. 7, pp. 19–34, Apr. 2016. C OMPUTATION, the IEEE T RANSACTIONS ON C YBERNETICS , and the IEEE
[118] H. Eskandar, A. Sadollah, A. Bahreininejad, and M. Hamdi, “Water A CCESS . His research interests span the areas of meta-heuristics, evolu-
cycle algorithm – a novel Metaheuristic optimization method for solv- tionary computation, swarm intelligence, cooperative algorithms, continuous
ing constrained engineering optimization problems,” Comput. Struct., optimization, large-scale optimization, Internet-of-Things (IoT), smart cities,
vols. 110–111, pp. 151–166, Nov. 2012. control & robotics, and engineering education.
[119] K. Gao, Y. Zhang, Y. Zhang, R. Su, and P. N. Suganthan, “Meta-
heuristics for bi-objective urban traffic light scheduling problems,”
IEEE Trans. Intell. Transp. Syst., vol. 20, no. 7, pp. 2618–2629,
Jul. 2019.
[120] K. Gao, Y. Zhang, R. Su, F. Yang, P. N. Suganthan, and M. Zhou, Mounib Khanafer (Senior Member, IEEE) received the B.Sc. degree in
“Solving traffic signal scheduling problems in heterogeneous traffic electrical engineering from Kuwait University, Kuwait, in 2002, the M.Sc.
degree in electrical engineering from the University of Ottawa, Canada, in
network by using meta-heuristics,” IEEE Trans. Intell. Transp. Syst.,
vol. 20, no. 9, pp. 3272–3282, Sep. 2019. 2007, and the Ph.D. degree in electrical and computer engineering from
[121] R. García-Ródenas, M. L. López-García, M. T. Sánchez-Rico, and the University of Ottawa, Canada, in 2012. He is currently an Associate
Professor of electrical and computer engineering with the Department of
J. A. López-Gómez, “A bilevel approach to enhance prefixed traffic
signal optimization,” Eng. Appl. Artif. Intell., vol. 84, pp. 51–65, Engineering, American University of Kuwait, Kuwait. He conducts research
Sep. 2019. in the general areas of wireless sensor networks and the Internet of Things
[122] A. Jamal, M. Tauhidur Rahman, H. M. Al-Ahmadi, I. Ullah, and and their applications. He has three years of industrial experience at Nortel
Networks, Canada, and worked as a Post-Doctoral Fellow with the School of
M. Zahid, “Intelligent intersection control for delay optimization:
Using meta-heuristic search algorithms,” Sustainability, vol. 12, no. 5, Electrical Engineering and Computer Science, University of Ottawa, Canada,
p. 1896, Mar. 2020. from 2012 to 2013.
[123] C. G. Corlu, R. de la Torre, A. Serrano-Hernandez, A. A. Juan, and
J. Faulin, “Optimizing energy consumption in transportation: Literature
review, insights, and research opportunities,” Energies, vol. 13, no. 5,
p. 1115, Mar. 2020. Kaizhou Gao (Member, IEEE) received the B.Sc. and master’s degrees,
[124] Y. Zhang, Y. Zhang, and R. Su, “Pedestrian-safety-aware traf- China, in 2005 and 2008, respectively, and the Ph.D. degree from Nanyang
fic light control strategy for urban traffic congestion alleviation,” Technological University (NTU), Singapore, in 2016. From 2008 to 2012, he
IEEE Trans. Intell. Transp. Syst., early access, Dec. 4, 2019, doi: was with the School of Computer, Liaocheng University, China. He was a
10.1109/TITS.2019.2955752. Research Associate with the School of Electronic and Electrical Engineering,
[125] G. Wu, R. Mallipeddi, and P. N. Suganthan, “Ensemble strategies NTU, from 2012 to 2013. From 2013 to 2015, he was a software engineer with
for population-based optimization algorithms—A survey,” Swarm Evol. the Singapore Institute of Manufacturing Technology, A*STAR, Singapore.
Comput., vol. 44, pp. 695–711, Feb. 2019. Since 2015, he has been a Research Fellow with the School of Electronic
[126] X. Ma et al., “A survey on cooperative co-evolutionary algorithms,” and Electrical Engineering, NTU. His research interests include intelligent
IEEE Trans. Evol. Comput., vol. 23, no. 3, pp. 421–441, Jun. 2019. computation, optimization, scheduling, and intelligent transportation. He has
[127] Y. Tan and K. Ding, “A survey on GPU-based implementation of authored over 70 refereed papers and is an Associate Editor of International
swarm intelligence algorithms,” IEEE Trans. Cybern., vol. 46, no. 9, Journals Swarm and Evolutionary Computation and IET Collaborative Intel-
pp. 2028–2041, Sep. 2016. ligent Manufacturing.

Authorized licensed use limited to: SASI INSTITUTE OF TECHNOLOGY & ENGINEERING. Downloaded on October 23,2024 at 05:35:24 UTC from IEEE Xplore. Restrictions apply.