Journal of Ambient Intelligence and Humanized Computing

https://doi.org/10.1007/s12652-018-0906-0

ORIGINAL RESEARCH

Ticket‑based QoS routing optimization using genetic algorithm

for WSN applications in smart grid
Uthman Baroudi1 · Manaf Bin‑Yahya2   · Meshaan Alshammari1 · Umair Yaqoub1

Received: 7 November 2017 / Accepted: 5 June 2018

© Springer-Verlag GmbH Germany, part of Springer Nature 2018

Abstract
Wireless sensor network (WSN) information network in Smart Grid is envisioned to handle diversified traffic such as real-
time sensitive data and non-real-time traffic. Therefore, QoS routing protocol in smart grid network is essential. Ticket-based
routing (TBR) protocol is a promising protocol because it can select routes based on several desired metrics, for example
route cost and delay. However, the original TBR suffers the need for transmitting a huge number of tickets to probe the sen-
sor network and discover the path cost and delay. Genetic algorithm can be used to minimize the number of tickets as well
as discovery messages overhead. In this work, we implement genetic algorithm (GA-TBR) at the source sensor node to col-
lect the state information inside the WSN environment of Smart Grid and hence optimize the selection of routes to ensure
the required QoS. Extensive simulation experiments have been conducted to investigate the performance of GA-TBR. The
simulation results have shown that with few tickets, the proposed algorithm is able to select routes with minimum possible
delay and shows 28% improvement compared to ad hoc on demand distance vector routing (AODV) protocol.

Keywords  Smart Grid · WSN · QoS Routing · Genetic Algorithm · Ticket-Based Routing · Route Discovery Optimization

1 Introduction Smart Grid is expected to replace the conventional power

grids in the coming years. In some parts of the world, they
SMART Grid has arisen as a new conception of next-gen- already form a prominent part of the power supply. Thus,
eration electricity grid for better efficiency and reliability different domains and elements are defined and specified in
through automated control and modern communications the new power system. The main domains are generation,
technologies. Smart Grid aims at enhancing the existing transmission, distribution, and customer domains (Gungor
electric power grid infrastructure by building an overlay et al. 2011). Figure 1 depicts an overall architecture of smart
communication and computing network that will increase grid; multiple sensors and actuators are distributed overall
the capacity and flexibility of the power grid (Yoldas et al. the smart grid. Moreover, these domains and elements can
2017). One of the main features of Smart Grid is that supply talk with each other in a large communication system to
and demand will be automatically controlled via two-way achieve the requirements of Smart Grid such as efficiency,
communication among different nodes. reliability, flexibility, and demand response. Furthermore,
Smart Grid attempts to benefit from the development of
Advanced Metering Infrastructure (AMI) as a smart meter
system, Wireless Sensor Network (WSN) as a monitoring
This study was funded by the deanship of scientific research at system and renewable energy systems (Tuballa and Abundo
King Fahd University of Petroleum and Minerals, under grant# 2016). AMI’s are smart meters that are more capable than
RG1319-1.
the conventional energy consumption counters. These smart
* Uthman Baroudi meters are installed at the electricity supplier as well as the
[email protected] location of the various consumers. Therefore, smart grid
poses several challenges to manage energy consumption
1
Department of Computer Engineering, King Fahd University and related costs in the side of smart homes such as profil-
of Petroleum and Minerals, Dhahran 31261, Saudi Arabia
ing (Gentile et al. 2016a, b), secure billing and financial
2
Department of Electrical and Computer Engineering,
University of Waterloo, Waterloo, ON N2L 3G1, Canada

13
Vol.:(0123456789)
 U. Baroudi et al.

Fig. 1  Smart grid architecture

(Gungor et al. 2011)

process (Yang et al. 2013), and online social networks effects and flexible solutions for providing sensing and control over
(Kamilaris et al. 2012). Smart Grid system. WSN features support low setup, main-
The effect of a failure of one node or zone in any of Smart tenance, and upgrading costs. Additionally, WSN function-
Grid domains can comprise the single failure segment, other ing has the advantage over the traditional systems in case
segments, or the entire system (Rekik et al. 2017). In the of self-organization, coverage, and online responses. The
current power systems, monitoring systems engage in lim- WSN monitoring system can be established across all Smart
ited assessments do not provide any automated handling of Grid domains. For instance, WSN can monitor power qual-
monitoring data where monitoring devices do not intercon- ity, renewable energy farms, and distributed generation at
nect with each other and provide a finite collection of data. the generation domain. In addition, detecting any failure in
The sensing system can monitor electrical and non-electri- transmission and distribution domains can improve Smart
cal parameters. For example, the monitored parameters for Grid stability. This failure can occur as a power outage or
transmission line are weather conditions and wire state to in overhead or underground transmission lines for several
determine the safety level of transmission operation. Fur- reasons such as overheating, natural disasters, animals, theft,
thermore, adding more electrical and mechanical parameters etc (Fadel et al. 2015).
can support overloading control by determining the dynamic On the other hand, several challenges raised from utiliz-
line rating (DLR). Also, providing electrical parameters such ing WSN in Smart Grid such as security, quality of service
as current and voltage can help for decisions in real time that (QoS) provisioning and transmission overhead. WSN system
improves the power system efficiency (Mazur et al. 2017). encounters many security threats due to its open environ-
Several communications technologies do exist for data ment (Chhaya et al. 2017). While solutions to overcome its
exchange among monitoring and control components vulnerabilities are not feasible due to its limited resources.
through wired and/or wireless media. Wired communica- Different solutions are proposed to solve security issues
tion is the most media used in the existing power systems of WSN in Smart Grid with specific design requirements.
such as Ethernet and fiber optic (Fateh et al. 2013). Wireless Moreover, reliable communication with minimum latency
networks provide a promising solution with the develop- and adequate bandwidth (Ancillotti et al. 2013) is required
ment of WSN due to inexpensive infrastructure and ease to deliver real-time information inside Smart Grid and it
of deployment for unreachable and isolated areas. ZigBee should provide QoS depending on the type of data trans-
(Fadel et al. 2015) is one of these wireless technologies that mitted through the WSN network. QoS routing protocol in
provide low cost and ease of setup and implementation with WSN-Smart Grid networks must select paths depending on
addition of low power usage. several QoS metrics to achieve certain requirements. How-
Summing up, WSN can provide vital monitoring data ever, obtaining best paths to meet QoS constraints requires
that is used in Smart Gird assessment besides automated many control messages, which will increase the background
decisions. In addition, deploying WSN guarantees reliable traffic. Hence, delay and congestion in data network will

13
Ticket‑based QoS routing optimization using genetic algorithm for WSN applications in smart…

grow as routing overhead increases. For this reason, route – We have compared our proposed approach GA-TBR with
discovery background traffic must be reduced. AODV which is adopted by IEEE 802.11s and GA-TBR
Ticket-based routing (TBR) (Xiao et al. 2002) is a well- shows 28% improvement on average.
known method for finding optimal routes in ad-hoc net-
works. It is applied by broadcasting multiple probes from In addition, the proposed algorithm can lend itself to other
the source to a destination via multiple routes to find the applications such as social networking based smart grid
best path. However, the number of sensor nodes is expected where each social group can constitute one node in the whole
to increase considerably and link states keep changing due network graph (Huang et al. 2015). For example, users can
to mobility and sporadic nature of the connection. There- exploit their social network to facilitate sending grid outage
fore, TBR becomes a highly complicated, cost-ineffective alerts notification and visualization, repair work schedules as
process (Levis et al. 2009). As the amount of information well as brand building (Moreno-Munoz et al. 2016).
transferred is expected to increase significantly in the near This paper structure is organized as follows. Section 2
future with services like video surveillance that require high summarizes the existing work about QoS protocols in Smart
QoS, a more efficient approach is required. Therefore, it is Grid and researches that use genetic algorithm for routing
very important to design a robust efficient routing algorithm optimization. Section 3 starts by describing the TBR proto-
that is capable of providing high QoS to Smart Grid net- col in 3.1 along with a detailed example in 3.2. The proposed
works traffic. genetic based algorithm is described in details in section 3.3.
Recently, route discovery optimization in ad hoc net- Section  4 presents the simulation models and discusses
works environment is given a significant attention using the results. Then, analysis of our proposed genetic based
metaheuristic algorithms such as genetic algorithm. There approach and the performance evaluation of TBR along with
are numerous research works that use genetic algorithm in the genetic algorithm are provided in Sect. 5. Finally, Sect. 6
order to shrink routing overhead in data networks as (Yen concludes the paper and provides suggestions for the future
et al. 2011; Zafar and Soni 2014; Cheng and Yang 2010b). work.
The main motivation for such direction is the great potential
of metaheuristic approaches to find near-optimal solution in
reasonable running time time. This feature is very desirable 2 Related works
in industrial applications. In this work, we follow the same
path and develop a genetic algorithm to optimize route dis- Saputro et al. 2012 have provided a survey of the exist-
covery overhead using TBR protocol. At the source node, ing routing protocols in the Smart Grid communications
the genetic algorithm is applied to find optimal paths that infrastructure. They extensively studied and analyzed the
satisfy the QoS requirement. In order to discover the optimal advantages and disadvantages of the proposed protocols with
paths, paths from route replies are used as initial population respect to different purposed areas. They argued that there
and the fitness function depends on the required QoS met- is not enough research on QoS routing in the Smart Grid
rics. In this paper, we use delay as QoS metric just to show networks. In addition, different QoS requirements of WSN
an example. However, using TBR protocol, any quantita- applications in smart grids must be addressed such as reli-
tive QoS metric which can be accumulated could be used ability and latency (Gungor et al. 2013). Nevertheless, there
as a fitness function such as throughput and delay jitter. The are several surveys studied the impact of WSN deployment
main contributions of this work can be summarized in the in Smart Grids (Rekik et al. 2017; Fadel et al. 2015; Wang
following points: et al. 2015; Usman and Shami 2013). Current researches
show that the development of WSN pushes towards address-
– Proposing an enhanced GA-TBR algorithm that can ing monitoring and control challenges in the new power sys-
select routes according to a predefined set of QoS of tems and provide an immense potential to satisfy its commu-
requirements with minimal probing tickets and compu- nication requirements. In the rest of this section, we provide
tational complexity. a brief literature review about existing WSN applications
– Exploiting the routing cache in every sensor node to in Smart Grid networks and its challenges. Also, the QoS
construct a connectivity matrix that will be employed communication issues and possible generic algorithms based
to validate the generated offsprings. This feature is very solution are discussed for such environments.
essential part of our algorithm that reduces the overhead Fateh et al. (2013) proposed a hybrid wireless structure for
messages as well as the required number of tickets. the transmission line monitoring. The hierarchical architec-
– Performing an in-depth analysis of certain features of ture aims to reduce the setup and functioning costs without
genetic algorithm such as validity checking and the fit- dimming the communication requirements such as delay and
ness function. bandwidth. However, using cellular nodes to relay data to the
control centers raise the cost of this network design. However,

13
 U. Baroudi et al.

there is a tradeoff between network cost and transmission delay to improve routing performance but not in the context of
(Li et al. 2016). Therefore, Li et al. (2016) studied the number QoS (Ahn and Ramakrishna 2002). However, an efficient
of cellular modules and their positions in order to achieve low method to check the validity has still not been specified.
network cost associated with low transmission delay. Furthermore, the methods described to discover crossovers
Kovendan and Sridharan (2016) proposed a WSN imple- and mutations are ineffective because they require longer
mentation in a Smart Grid model that provides a monitor- convergence time. Furthermore, genetic algorithm is used
ing of transmission and generation side. Kurt et al. (2017) to solve dynamic shortest path routing problem in mobile
proposed a novel method to calculate the optimal packet size ad hoc network (Yang et al. 2010), wireless mesh network
of WSN for Smart Grid applications. This technique tries to (Jiang et al. 2010). Immigrants and memory schemes are
specify real-time channel characterization in order to obtain proposed to enhance genetic algorithm for dynamic QoS
packet size optimization measurements for WSN in Smart routing problem (Cheng and Yang 2010a). Jiang et al. (2010)
Grid. He et al. (2017) studied security mechanisms for WSN proposed QoS routing optimization problem solution for
applications of Smart Grid. Authors analyze cyber security wireless mesh networks using hybrid genetic algorithm and
risks that such environments are vulnerable to by provid- ant colony optimization algorithm.
ing security assessment and vulnerabilities classification.
While in (Yan et al. 2017), a watermarking-based security
model is proposed to attain security requirements with low 3 Genetic ticket‑based routing protocol
operational cost.
Vallejo et al. (2012) described how a QoS broker can be In order to have a better understanding of our proposed
used to enhance the QoS of smart grid communication by approach, we start first with introducing the main concepts
providing QoS management in a centralized and standard- of TBR along with a detailed example. Then, we describe
ized manner to meet their strict requirements. QoS broker the genetic ticket-based routing protocol (GA-TBR).
devices can update certain parameters of layer 3 and layer
2 networks to improve the effectiveness of end to end QoS 3.1 Ticket‑Based Routing Protocol
through Smart Grid communication. Li and Zhang (2010)
proposed optimized multi-constrained routing (OMCR) pro- Ticket-based routing (TBR) (Chen and Nahrstedt 1998)
tocol which is a greedy algorithm implemented for secure is an on-demand routing protocol such as dynamic source
QoS routing protocol and satisfy real-time system require- Routing (DSR) and ad hoc on demand distance vector rout-
ment that can handle the impact of communication metrics. ing (AODV). On-demand routing protocol does not keep
Two QoS parameters namely delay and outage probability updated route tables with the most recent route topology.
are used in this multi-constrained QoS routing protocol. The Enhanced ticket based routing (ETBR) protocol (Xiao et al.
authors assumed that a home appliance can communicate 2002) improved the searching ability of discovery method
with the control center by sending a QoS requirement and in original TBR. However, the number of tickets used to find
then the control center assigns one or more routes for the a route is still the same. In other words, ETBR causes the
home appliance to guarantee the QoS requirement. same control message overhead as original TBR protocol.
On the other hand, the genetic algorithm can be used in In TBR, when a sensor wants to send data, it should dis-
ad-hoc network protocols for optimizing the route discov- cover a route based on certain metrics like cost and delay.
ery process (Sara and Sridharan 2014). A multi-cast tree This sensor starts sending probes with a certain number of
construction for QoS routing mechanism for mobile ad hoc tickets to all neighbor sensor nodes. In contrast to DSR,
network is presented in (Lu and Zhu 2013) to moderate end- intermediate sensor nodes do not flood these probes to its
to-end delay and total energy cost using genetic algorithm. neighbors; each sensor distributes tickets depending on a
A genetic algorithm is applied on delay constrained source distribution method defined by many parameters such as the
based mechanism to reduce route selection power consump- number of tickets received and historical probes information
tion as well as delay. Zaballos et al. (2013) proposed a new records. In addition, each sensor node adds its address to
QoS protocol based on TBR that uses genetic algorithm for (source to destination) path and accumulates the path cost.
minimizing route discovery traffic. Multi-paths are probed When a probe reaches the destination or a sensor node that
at the same time and randomly forwarded across the network have a route to the destination on its route cache, this node
to return paths that meet QoS constraints. The maximum sends a route reply (similar to PREP in DSR) containing
initial population is the number of tickets issues to estab- the source to destination path with its cumulative cost and
lish a connection. After the mutation operation, the node delay. Moreover, each sensor node has a route cache that
sends a probe with a single ticket to validate the mutated stores routes, which have been discovered before by sending
routes causing an increase in message overhead and longer probes or learned from route discovery messages of others as
delay. In contrast, genetic algorithms have been applied intermediate nodes. The sensor node instead of originating a

13
Ticket‑based QoS routing optimization using genetic algorithm for WSN applications in smart…

new request to discover route or forwarding others requests, tickets handled by this probe, and {path} denotes the accu-
it can use a route from its cache. Using historical informa- mulated path. Sensor 1 is the only neighbor of the source
tion probes records can eliminate redundant paths to reach sensor. Hence, sensor 0 forwards the probe P4,{0} to sensor 1
the destination in addition to eliminating path cycles. In this with all number of tickets. After receiving P4,{0} , this probe
case, intermediate nodes pass only one probe message from is split into three new probes and forwarded to its neighbors.
the same sender node with the same sequence number of When a probe reaches sensor 4, this sensor stops forwarding
the route request. probes as a result of having a path to sensor 7 in its route
Generally, once a sensor generates a probe, it can split it cache thus a reply R1{0,1,4,3,6,7} is generated and sent back to
into a number of probes and then distribute them to its neigh- sensor 0. At sensor 5, P1,{0,1,3} is received before P1,{0,1,2} ,
bor nodes. Furthermore, every probe can be divided into therefore, the probe which arrives later will be discarded
more probes as long as there is enough quantity of tickets on by this sensor to minimize probing overhead and eliminate
this probe. Each probe split and forwarded until it reaches infinite cycles probing. Later, P1,{0,1,3,5} probe arrives the
the destination unless it gets dropped by an intermediate destination sensor within delay constraint; sensor 7 sends
node due to QoS constraints or optimizing reasons. Each a route reply R2{0,1,3,5,7} that contains the accumulated cost
probe reaches the destination can obtain possible routes. and delay to the source sensor. Finally, two feasible routes
Then, the destination node can handle only routes which are discovered from source and destination {0, 1, 3, 5, 7}
satisfy certain QoS requirements. For example, only paths and {0, 1, 4, 3, 6, 7} . Now, {0, 1, 3, 5, 7} and {0, 1, 4, 3, 6, 7}
with an accumulated delay less than certain delay threshold are used by GA as initial population. Let’s assume that gene
are considered as feasible paths. Since probing process will 3 is randomly selected as a crossing point. A new path of
increase messages load inside the network and consume a {0, 1, 3, 6, 7} results from the crossover operator and no
lot of bandwidth, further optimization is needed. We pro- validity check is required because the crossover is applied
pose to use genetic algorithm (described in Section 3.3) at to a common gene. This new resulted path can be an optimal
the source node and minimize the number of tickets. In this route from source to destination.
work, we assume that every sensor has its own neighbor list However, the number of probes and their tickets can still
obtained by link level protocol. Furthermore, any changes be minimized and hence the overall discovery messages
in topology such as the occurrence of a new sensor/actuator overhead. In other words, an optimal route can be discov-
or losing connection with a sensor node should be updated ered using genetic algorithm even with reduced number of
within a finite time. tickets (N).

3.2 Detailed TBR example 3.3 The proposed algorithm

Now, we will provide a detailed example of TBR route dis- Genetic algorithm (Goldberg 1989) is a search algorithm for
covery. Let assume a network of 8 sensors shown in Fig. 2 optimization, which is used to find a near-optimal solution
where sensor 0 wants to discover a route to sensor 7. First, through a search space using operations inspired by natural
sensor 0 initiates a probe with a certain finite number of genetics. Genetic algorithm improves the quality of its initial
tickets ( N = 4 ). Pn,{path} is a general notation for a probe in population and produces a new population of high-quality
this paper, where P denotes Probe, n denotes the number of outcomes where each outcome can solve the problem and the
outcome with the highest factor value is the optimal solution.
This algorithm has been proven theoretically and empiri-
cally to be effective in complex spaces by providing efficient
searching mechanism (Dengiz et al. 1997; Krasnogor and
Smith 2005; Liu et al. 2018). Nevertheless, it is not guaran-
teed that the produced offsprings are always optimal routes;
but it is acceptable given the reduced computational cost.
In our proposed method, the genetic algorithm is used to
optimize the route discovery using TBR. At the source sen-
sor, the genetic algorithm is applied to find optimal paths
that satisfy the QoS requirement. We achieve this goal by
using paths constructed from probe tickets’ replies as initial
population. Each possible path from a probe ticket reply rep-
resents a chromosome, which consists of a number of genes.
Then, the fitness function is set to evaluate each chromo-
Fig. 2  Example of TBR route discovery in 8-sensor nodes network some from the population.

13
 U. Baroudi et al.

3.3.1 Initial population generation algorithm using crossover operator by partially swapping their
genes around a common gene unless swapping happens
In this initial stage, original TBR is employed in order to around randomly, it then picks crossing points. When the
probe and collect information about network connectivity crossover happens around a common gene, no validity
and other credentials such as delay, bandwidth, congestion, checking is required.
etc. Algorithm 1 depicts the pseudo code for this stage. More – Mutation: in this operation, mutation operator mutates
details are provided in the simulation setup section. the new offspring’s by randomly altering the value of

Algorithm 1 Initial Population Generation Algorithm

Result: Connectivity Matrix and set of paths
Initialization: Given the i node with j out branches, N tickets need to be sent
while Total tickets < N do
For n = 1 : j branches
send a random k probes [0, N ]
For each probe ticket
initialize a counter for each QoS requirement
if intermediate sensor then
For each probe ticket
update the traversed path
update counter for each QoS requirement
if QoS requirement Not satisfied then
drop probe
else
Send to the next node towards the destination
end
else
end
if Destination node then
Send a unicast message to source node
else
It is a source node
Compile all data
end
end

3.3.2 Genetic algorithm randomly selected gene and then it is added to the current
population. Then, the fitness function is computed again
At each generation, the following three operations of the for these chromosomes.
algorithm are performed iteratively:
Algorithm 2 depicts the pseudo code for this stage. More
– Selection: chromosomes from the current population are details are provided in the simulation setup section.
selected according to their level of fitness value while
others are discarded.
– Crossover: in this operation, a new set of chromosomes
are generated from two chromosomes in the population

13
Ticket‑based QoS routing optimization using genetic algorithm for WSN applications in smart…

Algorithm 2 Genetic Ticket-Based Routing Algorithm

Result: Set of best chromosomes
Initialization: Initial Population
while fitness value is satisfied do
Selection process: a set of ”good”chromosomes is picked
Crossover process: pick a pair of two chromosomes
if random selection then
swap genes around a common gene
generate a new offsprings
check validity
else
swap genes around a common gene
generate a new offsprings
end
Mutation process: pick randomly a highly connected node to be a mutation point
if part of a current chromosome then
choose another node
else
generate a new offsprings
end
end

The best resultant chromosomes (paths) are returned by 4 Simulation setup

the genetic algorithm as a final outcome, when the optimiza-
tion condition is achieved. Crossover and mutation opera- 4.1 TBR simulation setup
tors produce a new set of approximations and breed them
together exactly like what happens on genetic process in We use Network Simulator-2 (NS-2) (McCanne et al. 1997)
nature. Furthermore, the fitness function is determined based as our platform to simulate TBR as described in (ELMO
on the factors that concern our problem. In this work, the
cost and delay as QoS requirements specify the fitness func-
tion of the genetic algorithm.
In our example shown in Fig.  2, the source sensor
received two route replies {0, 1, 3, 5, 7} and {0, 1, 4, 3, 6, 7} .
In this case, these two routes can be used as initial popula-
tion of genetic algorithm integrated in source sensor to find
the optimal route that satisfies QoS requirements. Each path
represents a chromosome; each sensor in the chromosome
represents a gene. Hence, the number of genes in the chro-
mosome determine its size. For example, path {0, 1, 3, 5, 7}
is a chromosome consist of 5 genes.

Fig. 3  Number of probing messages per connection request (mes-

sages overhead) versus number of tickets (N)

Table 1  Simulation parameters Table 2  Number of probing messages per connection (message over-

Parameter Value head)
Number of tickets Number of probing Number of probing
Simulation period 1000 s
messages messages per connec-
Topology dimension 150 m × 150 m tion
Number of nodes 100
MAC type 802.15.4 10 15131 30.262
Protocol TBR 25 23042 46.084
Coverage range 35 m 50 30149 60.298
100 36670 73.34
1000 65787 131.574

13
 U. Baroudi et al.

– First, when a probe returns back to the sensor that

originates this probe.
– Second, when a probe passed again the same sensor.
– Third, when another probe from the same source
with the same request ID (sequence number) has
already passed this sensor.
Fig. 4  The simulation methodology; generating initial population – Finally, if the accumulative delay exceeds the con-
using NS-2 and then it is fed to Matlab to execute the genetic algo-
rithm and produce the best possible paths straint delay or any other QoS constraints.

The simulations are accomplished using NS-2 version 2.34

installed in Ubuntu environment. The network model used
in our simulation is randomly created. In 150m x 150m area,
one hundred sensor nodes are placed with zero mobility. The
transmission range is set to 35m for each sensor node. The
delay of each link is modeled as a uniform random variable
in the range of [0, 1] time unit. Table 1 summarizes the
simulation parameters. This scenario is very close to the
expected scenario in Smart Grid ad hoc networks environ-
ment where short-range wireless smart meters are deployed
in a home-area-network.
We run the simulation 5 times for each number of tickets
(i.e. N = 10, 25, 50, 100, and 1000). In each simulation
run, the source sensor node is fixed and it transmits routing
requests to 99 randomly selected destinations. Thus, we will
have for each case (different ticket number experiment) 495
Fig. 5  Connectivity matrix of 8 nodes network example shown in samples, which are statistically sufficient to achieve 95%
Fig. 2; ’0’ entry means no connectivity, ’1’ means entry means con- confidence in our collected initial population.
nectivity exits The average path cost and success ratio of TBR have been
studied in (Xiao et al. 2002). TBR has a lower ratio of all
routes cost to the number of routes (average path cost) than
2010). In the following, we describe some features of our flooding and shortest path algorithms. Moreover, the ratio of
implementation of TBR in NS-2: paths established to the total of route requests (success ratio)
was close to the flooding algorithm (Chen and Nahrstedt
– When the source needs to start probing or an interme- 1998).
diate sensor has to forward a certain number of tickets In order to demonstrate that our proposed approach has
(N) to its neighbors’ sensors, the number of tickets that better searching capability in addition to lower network load,
will be sent is determined randomly between [0, N] using we define message overhead metric for the routing discovery
uniform distribution routine. method. Message overhead metric is defined as the aver-
– When a probe is received by an intermediate sensor, the age number of probing messages required per connection
following actions are triggered. First, the accumulative request. Any probe passes a link from one sensor node to any
delay and cost are calculated. Second, the probe informa- neighboring sensor node is considered as one probe mes-
tion is stored a cache and it will be used later as a histori- sage. Therefore, when a probe passes k hops, it is recorded as
cal probing record. Finally, current sensor ID is added to k messages. Figure 3 depicts the message overhead for TBR
TBR header (i.e. traversed path) of the newly generated with different number of tickets. Clearly, we can observe the
probes delivered to its neighbors. larger the number of tickets, the larger is message overhead
– When a probe is received by the destination sensor, a uni- which also shown in Table 2.
cast route reply is sent back to the source using the same The values of N = 10, 25 and 50 cases are promising,
path after checking the QoS constraint as mentioned but we should not miss the success ratio of these cases.
before. However, for N = 25 and 50 cases, it can save 27.3% and
– Intermediate sensors will drop a probe in the following 17.8% of message overhead, respectively and their success
cases: ratios with respect to N = 100 case are about 71% and 91%,
respectively.

13
Ticket‑based QoS routing optimization using genetic algorithm for WSN applications in smart…

4.2 GA‑TBR simulation setup mutation discoveries in addition to a defined source and

destination for each particular run of the algorithm.
In this work, MATLAB is used along with NS-2 as our
simulation platforms to test the impact of genetic algorithm – Fitness Function: The fitness function depends on QoS
so that the performance of the algorithm can be analyzed routing metrics and the feasible route population is
deeply at each stage. The initial population of route replies selected accordingly. It will be sorted in a decreasing or
is exported from NS-2 as csv files. Figure 4 illustrates the non-decreasing order based on our requirement. Combi-
simulation steps used in this work. nation of more than one QoS metric can be achieved by
Moreover, the connectivity, cost and delay information assigning different weights for each metric.
of network model is exported from NS-2 in a matrix repre- – Validity Checking: We will exploit the information
sentation. Figure 5 shows the corresponding connectivity learned and cached in each sensor node to check the
matrix of the network described in Fig. 2. Where ’0’ value validity of the new routes instead of sending unicast mes-
indicates that there is no connection between these two sen- sages which will cause a huge background traffic. This
sor nodes. In other words, each sensor node that does not step will minimize the message overhead, alleviates the
exist within the coverage range of the other sensor node has network from extra load and reduces the network conges-
weight of ’0’. However, ’1’ value represents a connectivity tion.
between sensor nodes. In a similar manner, other matrices – Crossover: Two chromosomes will be selected for cross-
(i.e. delay, cost) are represented with appropriate weights over; if there is a common sensor node, it will be the
between sensor nodes. crossover point otherwise, it will be chosen randomly.
We have implemented five features in our genetic algo- A common sensor node will ensure validity, which will
rithm: fitness function, validity checking, crossover and save algorithm running time. For example, {0, 1, 3, 5, 7}
and {0, 1, 4, 3, 6, 7} are routes discovered by TBR and
GA uses them as an initial population. Let’s assume that
sensor node 3 is randomly selected as a crossing point.
A new path of {0, 1, 3, 6, 7} results from the crossover
operator and no validity check is required because the
crossover is applied to a common gene. This new resulted
path can be an optimal route from source to destination.
– Mutation: The algorithm will identify sensor nodes with
high connections and one of them will be chosen ran-
domly to be the mutation point. In order to enhance the
validity, the new sensor node should not be part of the
current chromosome.
– Source and Destination: Source and destination informa-
tion is added to check the validity. There is a possibility
that we might get a chromosome that does not have the
source sensor node, destination sensor node nor both. In
this case, we assign its fitness function to infinity to avoid
Fig. 6  Offspring evaluation after running the enhanced algorithm for selecting them as candidate solutions, when in reality
fitness and validity they are not.

Table 3  Statistical analysis for Random selection with Sequential selection with Sequential selection
three crossover techniques cut vertex cross over cut vertex cross over w/o cut vertex cross
over

Runs 100 100 100

Mean 0.909 0.747 0.770
STDEV 0.0933 0.0815 0.0838
Level of significance 0.05 0.05 0.05
Confidence level (95%) 0.182 0.159 0.163
Lower bound 0.727 0.588 0.607
Upper bound 1.091 0.905 0.934

13
 U. Baroudi et al.

Table 4  Statistical analysis for mutation performed on different popu-

lation percentages using random selection crossover with cut vertices
10% 20% 80% 90% 100%

Count 100 100 100 100 100

Mean 0.0808 0.0460 0.0268 0.0216 0.0157
STDEV 0.0116 0.0082 0.0087 0.0081 0.0034
Confidence Coefficient 0.95 0.95 0.95 0.95 0.95
Level of Significance 0.05 0.05 0.05 0.05 0.05
Confidence Level 0.0227 0.0161 0.0170 0.0158 0.0067
(95%)
Lower bound 0.0581 0.0298 0.0098 0.0058 0.0090
Upper bound 0.1035 0.0621 0.0437 0.0374 0.0225
Fig. 7  Comparison of average running time for different crossover
techniques with different selection methods performed on whole pop-
ulation
initial population with valid routes. Moreover, the algorithm
It is well known that the number of rows in the population needs a space (Population size) to add off springs; in this
matrix represents the number of candidate solutions (routes). case, the source sensor node is allowed to duplicate the ini-
On the other hand, the relationship between transmission tial population so it has valid routes with double size and the
range and the population matrix is the number of columns in random search will take care of duplicates. Therefore, a valid
the population matrix. In other words, the maximum trans- route most likely will lead to another valid route. In contrary,
mission range determines the chromosome length. Further- an invalid route most likely will lead to another invalid route
more, using the whole range is impractical for this type of or it will take much time to be validated by the algorithm.
optimization, so the route can be terminated after reaching Figure 6 presents our GA implementation by applying a cost
the destination sensor node using a special termination char- of infinity to sensor nodes where a connection does not exist.
acter (-1 in our case) to fill the remaining cells. This is useful To illustrate, when the algorithm is run, there is no feasible
in managing the maximum transmission range by restricting path between sensor nodes 1, 2, 4, and 8.
the number of columns in the population matrix.
One of the main advantages of genetic algorithms is ran-
domness; this implies that having a random initial popula- 5 Performance analysis
tion will be an advantage. However, this is not the case in
path optimization problems as we are looking for a valid Having introduced above the main features of our proposed
path and not any path. TBR can provide valid paths so the approach, we shall first analyze the genetic algorithm alone
genetic algorithm can manipulate them and enhance the and then we evaluate the performance of TBR along with
validity. Even if a random initial population has a large size, the genetic algorithm.
it does not have the same impact on the algorithm as a small

Fig. 8  100 runs show the behav-

ior of three crossover techniques
with respect to time

13
Ticket‑based QoS routing optimization using genetic algorithm for WSN applications in smart…

Table 5  Run time (in seconds) results of cost and delay as QoS metrics with various numbers of tickets (N) of TBR
Number of tickets Cost Delay Overall
(N) MEAN MAX MIN MEAN MAX MIN MEAN MAX MIN

10 0.035025 0.09311 0.000759 0.025254 0.090229 0.000622 0.03014 0.09311 0.000622

25 0.033249 0.09434 0.000762 0.02052 0.070107 0.000548 0.026885 0.09434 0.000548
50 0.03069 0.087502 0.000597 0.019295 0.099336 0.000421 0.024993 0.099336 0.000421
100 0.029454 0.092047 0.000617 0.018653 0.0996 0.000435 0.024053 0.0996 0.000435
1000 0.027749 0.09891 0.000459 0.015425 0.099041 0.000419 0.021587 0.099041 0.000419
Overall 0.029563 0.09891 0.000459 0.017785 0.0996 0.000419 0.023674 0.0996 0.000419

Table 6  Run time (in seconds) results comparison with various number of paths in initial population of genetic algorithm
Initial Population Cost Delay Overall
MEAN MAX MIN MEAN MAX MIN MEAN MAX MIN

2 or 3 0.028696 0.09891 0.000459 0.016201 0.099041 0.000419 0.022448 0.099041 0.000419

4 or more 0.02164 0.080992 0.000751 0.01476 0.084966 0.000564 0.0182 0.084966 0.000564

5.1 Genetic algorithm analysis

It is important to perform the crossover and mutation prob-

ability analysis and compare the performance among dif-
ferent possible approaches for route optimization. As men-
tioned before, the crossover is one of the main components
of genetic algorithms where pairs of chromosomes will be
selected and a crossing point will be set by the algorithm.
Then, the pairs of offspring will be produced. Selecting the
mid of the chromosomes as crossover points will most likely
produce invalid routes and it increases the overhead on the
algorithm due to invalid routes. Hence, it is highly neces-
sary to choose the crossover point carefully. Looking for
Fig. 9  Run time results (y-axis) of cost, delay and overall (both cost
cut vertices in the selected chromosomes will enhance the and delay) as QoS metrics with various numbers of tickets (x-axis)
validity of the produced offspring, which can save the run- of TBR
ning time. The crossover phase will look for the common
sensor node (cut vertex), otherwise, it will be the mid of the
chromosome.
The selection phase has been implemented by two meth-
ods namely, sequential and random selection. In the sequen- shows the shortest running time. The stability of the algo-
tial method, the pairs will be selected as they appear in the rithm with respect to time is a critical factor for genetic algo-
population. On the other hand, the random selection will rithms to hit the optimal solution; Fig. 8 shows the behavior
choose a chromosome randomly and cross it over with the of three crossover techniques with respect to time with 100
following chromosome. Furthermore, in order to investigate runs. Again, we can observe that the random selection with
the impact of cut-vertex sensor nodes, we have tested the crossover has the least fluctuations, while the sequential
following methods: Sequential Selection Crossover without selection without crossover has shown the highest incident
Cut Vertices, Sequential Selection Crossover with Cut Ver- of fluctuations. These fluctuations affect the stability of the
tices and Random Selection Crossover with Cut Vertices. genetic algorithm. Therefore, we pick the random selection
Table 3 summarizes the comparison between the crossovers with cross over technique for testing the mutation process.
using cut vertices and using mid of the chromosome. Fig- The algorithm will identify the connected sensor nodes
ure 7 shows the average running time of different crossover and one of them will be chosen randomly to be the muta-
techniques with different selection methods performed on tion point. The mutation point connections will be imported
the whole population. The random selection with cross-over from the cost matrix; the connections will be shuffled and

13
 U. Baroudi et al.

the comparison between different percentages of mutation

candidate chromosomes; we have tested them with Random
Selection Crossover with Cut Vertices.
As explained above, source and destination are added to
check the validity. There is a possibility that the algorithm
produces a chromosome that does not contain the source
sensor node, destination sensor node or both. Moreover,
checking whether the source and destination sensor nodes
within the chromosome can be very helpful in case of par-
allelism. Assume an intermediate sensor node “A” tries to
figure out the optimal route to sensor node “D”; in order
to apply parallelism, sensor node “A” will call its neighbor
sensor nodes “B” and “C” to search for a route from “A” to
Fig. 10  Run time results (y-axis) comparison with various number of
paths in initial population of genetic algorithm
“D”; in this case, sensor node “A” has to pass current source
and destination information as parameters to sensor nodes
“B” and “C”. Finally, “A” will collect the results (i.e. from
B, C) along with its own result and evaluate them to choose
the best route.

5.2 Genetic algorithm with TBR analysis

Having established and studied the performance of the pro-

posed genetic algorithm (GA), we are ready to study the
integration of TBR with GA. As mentioned before and
shown in Fig. 4; In order to evaluate the genetic algorithm
overhead, we have used NS-2 to simulate TBR that pro-
duces matrices of connectivity, cost and delay as well as
discovered paths for each connection request considering
different number of tickets. These matrices of network model
Fig. 11  Cost metric comparison between AODV and GA-TBR proto-
col using a random topology of 100 nodes with fixed source and 10 information and discovered paths are saved as csv files. In
random destinations MATLAB, we use these csv files as inputs for running the
genetic algorithm code.
In Tables 5 and 6, we present minimum, maximum and
average execution time of genetic algorithm in Matlab for
paths and topology generated in NS-2. These results show
that the additional processing overhead of genetic algorithm
used in our proposed method to discover the best feasible
route is very minimal. Cost and delay are used as QoS metric
individually to obtain feasible routes. As shown in Fig. 9, the
computation time increases with the decrease of number of
tickets where the mean execution time of N = 1000 is about
(0.0296s - Cost, 0.0178s - Delay) and that for N = 25 is
about (0.03325s - Cost, 0.0205s - Delay). Delay metric has
lower minimum and average execution time due to delay
aware behavior of TBR routing to find low cost/delay paths.
Fig. 12  Delay metric comparison between AODV and GA-TBR pro- Furthermore, increasing number of tickets to 1000 surely
tocol using a random topology of 100 nodes with fixed source and 10 will decrease the overall delay, yet, it is not effective because
random destinations of the huge overhead traffic that will overwhelm the whole
network and cause congestion.
one of them is randomly chosen. To enhance the validity, the Table 6 presents the execution time which is needed based
new sensor node should not be part of the current chromo- on initial population size. The mean time required to deter-
some and the algorithm should not choose the destination mine the feasible path from 2 or 3 discovered paths is about
sensor node as a mutation point. In Table 4, we illustrate (0.0287s—cost, 0.0162s—delay). While, lower computation

13
Ticket‑based QoS routing optimization using genetic algorithm for WSN applications in smart…

time is needed for larger initial population where the mean Furthermore, although social network could help in pro-
time is about (0.0216s—cost, 0.0148s—delay) as shown in moting the vision of smart grid, incorporating this type of
Fig. 10. open networks would lead to misinformation attack problem
that could cause great damage to the smart grid (Pan et al.
5.3 Comparison of genetic algorithm with AODV 2017). As a future direction, we will investigate the perfor-
mance of GA-TBR under such attack.
Using the same topology composed of 100 nodes, we have
tested the performance of AODV using ns-2 simulator. Ten
cases have been examined where we have fixed the source References
node and randomly chosen ten destination nodes. Figures 11
and 12 depict the number of hop count and total delay for Ahn CW, Ramakrishna RS (2002) A genetic algorithm for shortest path
routing problem and the sizing of populations. IEEE Trans Evolut
each case, respectively. We can observe that GA-TBR Comput 6(6):566–579
outperforms AODV in all cases except case#8; it proves Ancillotti E, Bruno R, Conti M (2013) The role of communication sys-
the notion that have been discussed above that GA does tems in smart grids: architectures, technical solutions and research
not always provide the best answer. However, in all other challenges. Comput Commun 36(17):1665–1697
Chen S, Nahrstedt K (1998) Distributed qos routing with imprecise
cases GA-TBR shows outstanding performance with 28% state information. In: Computer Communications and Networks,
improvement on average. In addition, in some cases such as 1998. Proceedings. 7th International Conference on IEEE, pp
case#1 GA-TBR shows 68% improvement. 614–621
Cheng H, Yang S (2010a) Genetic algorithms with immigrants schemes
for dynamic multicast problems in mobile ad hoc networks. Eng
Appl Artif Intell 23(5):806–819
6 Conclusion Cheng H, Yang S (2010b) Multi-population genetic algorithms with
immigrants scheme for dynamic shortest path routing problems in
Smart Grid communications network is composed of a mobile ad hoc networks. In: Applications of Evolutionary Com-
putation, Springer, pp 562–571
numerous number of sensors and actuators. This network Chhaya L, Sharma P, Bhagwatikar G, Kumar A (2017) Wireless sensor
will carry diversified types of traffic, each has its own QoS network based smart grid communications: cyber attacks, intru-
requirements/constraints. Routing selection plays a key role sion detection system and topology control. Electronics 6(1):5
in satisfying these constraints. In this work, we proposed Dengiz B, Altiparmak F, Smith AE (1997) Local search genetic algo-
rithm for optimal design of reliable networks. IEEE Trans Evolut
a meta-heuristic technique for routing discovery of QoS Comput 1(3):179–188
aware TBR protocol in a WSN-Smart Grid environment. ELMO (2010) Ns2: How to add new routing protocol—causal nexus.
Genetic algorithm operations are used to explore new fea- http://elmur​od.net/en/index​.php/archi​ves/157
sible routes without sending an extra number of tickets. We Fadel E, Gungor VC, Nassef L, Akkari N, Malik MA, Almasri S, Aky-
ildiz IF (2015) A survey on wireless sensor networks for smart
validated mutated route using information stored in sensor grid. Comput Commun 71:22–33
nodes caches. Therefore, there are no extra routing messages Fateh B, Govindarasu M, Ajjarapu V (2013) Wireless network design
overhead required for validating new routes (off springs) due for transmission line monitoring in smart grid. IEEE Trans Smart
to unicast traffic; this is a huge advantage of our approach Grid 4(2):1076–1086
Gentile U, Marrone S, Mazzocca N, Nardone R (2016a) Cost-energy
compared to existing works. On the other hand, the results modelling and profiling of smart domestic grids. Int J Grid Util
of genetic algorithm running time demonstrated that our pro- Comput 7(4):257–271
posed method has minimum execution overhead. In addi- Gentile U, Marrone S, Mazzocca N, Nardone R (2016b) 𝜇grimoire: A
tion, GA-TBR shows 28% average improvement compared tool for smart micro grids modelling and energy profiling. The
Open Cybernetics & Systemics Journal 10(1)
to AODV and in some cases it shows 68% improvement. Goldberg DE (1989) Genetic algorithms in search, optimization, and
The proposed enhancement on TBR protocol was tested machine learning, 1989. Reading: Addison-Wesley
on a network model that has fixed sensor nodes. It is an Gungor VC, Sahin D, Kocak T, Ergut S, Buccella C, Cecati C, Hancke
interesting and challenging problem to test our approach on GP (2011) Smart grid technologies: communication technologies
and standards. IEEE Trans Ind Inf 7(4):529–539
mobile and dynamic environment. In addition, in this work, Gungor VC, Sahin D, Kocak T, Ergut S, Buccella C, Cecati C, Hancke
only the source sensor nodes applied the genetic algorithm, GP (2013) A survey on smart grid potential applications and com-
what if intermediate sensor nodes or even the destination munication requirements. IEEE Trans Ind Inf 9(1):28–42
sensor nodes would participate in this process? On the other He D, Chan S, Guizani M (2017) Cyber security analysis and protec-
tion of wireless sensor networks for smart grid monitoring. IEEE
hand, one of the limitations of this work is that we did not Wireless Communications
evaluate resource consumption of our proposed approach Huang Q, Li X, Zhao J, Wu D, Li XY (2015) Social networking reduces
such as power and memory. So further evaluation is needed peak power consumption in smart grid. IEEE Trans Smart Grid
to assess our approach due to the usage of heuristic algo- 6(3):1403–1413
Jiang H, Zheng L, Liu Y, Zhang M (2010) Multi-constratined qos
rithm which requires more processing and memory in the routing optimization of wireless mesh network based on hybrid
participate node.

13
 U. Baroudi et al.

genetic algorithm. In: Intelligent Computing and Integrated optimizations, and validation platforms. Wireless Personal Com-
Systems (ICISS), 2010 International Conference on, IEEE, pp munications pp 1–23
862–865 Saputro N, Akkaya K, Uludag S (2012) A survey of routing protocols
Kamilaris A, Taliadoros G, Pitsillides A, Papadiomidous D (2012) The for smart grid communications. Comput Netw 56(11):2742–2771
practice of online social networking of the physical world. Int J Sara GS, Sridharan D (2014) Routing in mobile wireless sensor net-
Space-Based Situat Comput 2(4):240–252 work: a survey. Telecommun Syst 57(1):51–79
Kovendan A, Sridharan D (2016) Wireless sensor networks based con- Tuballa ML, Abundo ML (2016) A review of the development of smart
trol strategies for the enhancement of reliability in smart grids. grid technologies. Renew Sustain Energy Rev 59:710–725
Circuits Syst 7(09):2499 Usman A, Shami SH (2013) Evolution of communication technolo-
Krasnogor N, Smith J (2005) A tutorial for competent memetic algo- gies for smart grid applications. Renew Sustain Energy Rev
rithms: model, taxonomy, and design issues. IEEE Trans Evolut 19:191–199
Comput 9(5):474–488 Vallejo A, Zaballos A, Selga JM, Dalmau J (2012) Next-generation qos
Kurt S, Yildiz HU, Yigit M, Tavli B, Gungor VC (2017) Packet size control architectures for distribution smart grid communication
optimization in wireless sensor networks for smart grid applica- networks. Commun Magazine, IEEE 50(5):128–134
tions. IEEE Trans Ind Electron 64(3):2392–2401 Wang P, Hou H, He X, Wang C, Xu T, Li Y (2015) Survey on applica-
Levis P, Tavakoli A, Dawson-Haggerty S (2009) Overview of existing tion of wireless sensor network in smart grid. Procedia Comput
routing protocols for low power and lossy networks. IETF Draft Sci 52:1212–1217
Report Xiao L, Wang J, Nahrstedt K (2002) The enhanced ticket-based routing
Li H, Zhang W (2010) Qos routing in smart grid. In: Global Telecom- algorithm. In: Communications, 2002. ICC 2002. IEEE Interna-
munications Conference (GLOBECOM 2010), 2010 IEEE, pp 1–6 tional Conference on, IEEE, vol 4, pp 2222–2226
Li M, Chi XB, Jia XC, Zhang JL (2016) Wsn-based efficient moni- Yan X, Zhang L, Wu Y, Luo Y, Zhang X (2017) Secure smart grid
toring for overhead transmission line in smart grid. In: Control communications and information integration based on digital
Conference (CCC), 2016 35th Chinese, IEEE, pp 8485–8489 watermarking in wireless sensor networks. Enterprise Inf Syst
Liu W, Gong M, Wang S, Ma L (2018) A two-level learning strategy 11(2):223–249
based memetic algorithm for enhancing community robustness of Yang J, Cheng R, Liu W, Xiao Y, Zhang F (2013) Cryptoanalysis and
networks. Inf Sci 422:290–304 improvement of smart prepayment meter protocol in standard q/
Lu T, Zhu J (2013) Genetic algorithm for energy-efficient qos multicast gdw 365. Int J Grid Utility Comput 4(1):40–46
routing. IEEE Commun Lett 17(1):31–34 Yang S, Cheng H, Wang F (2010) Genetic algorithms with immigrants
Mazur K, Wydra M, Ksiezopolski B (2017) Secure and time-aware and memory schemes for dynamic shortest path routing problems
communication of wireless sensors monitoring overhead transmis- in mobile ad hoc networks. IEEE Trans Syst, Man, Cybern, Part
sion lines. Sensors 17(7):1610 C (Appl Rev) 40(1):52–63
McCanne S, Floyd S, Fall K, Varadhan K, et al (1997) Network simu- Yen YS, Chao HC, Chang RS, Vasilakos A (2011) Flooding-limited
lator ns-2 and multi-constrained qos multicast routing based on the genetic
Moreno-Munoz A, Bellido-Outeirino F, Siano P, Gomez-Nieto M algorithm for manets. Math Comput Model 53(11):2238–2250
(2016) Mobile social media for smart grids customer engage- Yoldas Y, Onen A, Muyeen S, Vasilakos AV, Alan I (2017) Enhancing
ment: Emerging trends and challenges. Renewable and Sustain- smart grid with microgrids: challenges and opportunities. Renew
able Energy Reviews 53:1611–1616, https​://doi.org/10.1016/j. Sustain Energy Rev 72:205–214
rser.2015.09.077, http://www.scien​cedir​ect.com/scien​ce/artic​le/ Zaballos A, Vernet D, Selga JM (2013) A genetic qos-aware routing
pii/S1364​03211​50104​73 protocol for the smart electricity networks. International Journal
Pan T, Mishra S, Nguyen LN, Lee G, Kang J, Seo J, Thai MT (2017) of Distributed Sensor Networks 2013
Threat from being social: Vulnerability analysis of social network Zafar S, Soni M (2014) Trust based qos protocol (tbqp) using meta-
coupled smart grid. IEEE Access 5:16,774–16,783 heuristic genetic algorithm for optimizing and securing manet. In:
Rekik S, Baccour N, Jmaiel M, Drira K (2017) Wireless sensor net- Optimization, reliabilty, and information technology (ICROIT),
work based smart grid communications: Challenges, protocol 2014 International Conference on, IEEE, pp 173–177

13