Accepted Manuscript

A binary artificial bee colony algorithm for constructing spanning trees

in vehicular ad hoc networks

Xin Zhang, Xiu Zhang

PII: S1570-8705(16)30171-8
DOI: 10.1016/j.adhoc.2016.07.001
Reference: ADHOC 1414

To appear in: Ad Hoc Networks

Received date: 3 May 2016

Revised date: 21 June 2016
Accepted date: 3 July 2016

Please cite this article as: Xin Zhang, Xiu Zhang, A binary artificial bee colony algorithm for
constructing spanning trees in vehicular ad hoc networks, Ad Hoc Networks (2016), doi:
10.1016/j.adhoc.2016.07.001

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service
to our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and
all legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

A binary artificial bee colony algorithm for constructing spanning trees in

vehicular ad hoc networks

Xin Zhanga,b , Xiu Zhanga,b∗

a College of Electronic and Communication Engineering, Tianjin Normal University, Tianjin, China
b Tianjin Key Laboratory of Wireless Mobile Communications and Power Transmission, Tianjin Normal University, Tianjin, China

T
IP
Abstract
To accomplish reliable and efficient information routing, strong paths connecting all nodes are required in vehicular ad

CR
hoc networks (VANETs). Classical algorithms in graphic theory could find only one minimum spanning tree (MST) in
VANETs. Swarm intelligence paradigms are able to obtain several alternatives to MST, which is useful for improving
reliability of VANETs. This paper proposes a binary coded artificial bee colony (BABC) algorithm for tackling
the spanning tree construction problem. A two-element variation technique is designed to keep the consistence of

US
binary coded solutions. The proposed algorithm is applied to tackle a roadside-to-vehicle communication example.
The success rate and average hitting time of the algorithm to find MST are also analyzed. It is found that the BABC
algorithm could find MST with 92% probability. Though it is slower than Kruskal algorithm in terms of computational
time, the BABC algorithm can attain several suboptimal spanning trees in one run. This suggests that the algorithm
AN
would be useful under the condition that tree paths are required to be rebuilt frequently while the network topology is
unchanged in a short period.
Keywords: Artificial bee colony; vehicular ad hoc network; minimum spanning tree; binary representation.
M

1. Introduction
ED

There are two types of wireless networks — base-station oriented networks and ad hoc wireless networks. Hybrid
wireless network is formed by these two networks [1]. The ad hoc network topology is desirable because of its low
cost, plug-and-play convenience, and flexibility. Its usage of bandwidth and battery power is more efficient. Due
to the lack of infrastructure, the data is forwarded to the destination via a multi-hop fashion. Generally, the ad hoc
PT

network has been studied in optimization [2], target detection [3, 4], etc. In some scenarios, a set of base stations
are connected by wired links and placed within the ad hoc networks to form a wired infrastructure, intending to
enhance the whole network performance. This resulting network is referred to as a hybrid wireless network. Due to
CE

the dynamic nature of such network, quite often computational intelligence approaches such as fuzzy logic systems
[5, 6], swarm intelligence could be applied to optimization in hybrid wireless networks. This paper aims to apply
artificial bee colony (a recently created swarm intelligence paradigm) to vehicular ad hoc network (VANET).
Along with the development of intelligent transportation systems (ITSs), VANET becomes a major network com-
AC

munication topic in ITSs. VANET supports communication among arbitrary vehicles, which is realized by wireless
communication and data transmission techniques. The network organizes vehicles and roadside transport facilities
in order to let all customers receive and send real time traffic-related information. Moreover, VANET can enhance
traffic efficiency and also assure road safety and traveling comfort. VANET is an important branch of mobile ad
hoc network (MANET) with specific properties. In vehicular ad hoc networks (VANETs), vehicles are considered as
nodes. VANETs include roadside-to-vehicle communication (RVC), inter-vehicle communication (IVC), and hybrid
vehicular communication (HVC). Roadside unit equipment is usually deployed at service station, dangerous road

∗ Correspondingauthor.
Email addresses: [email protected] (Xin Zhanga,b ), [email protected] (Xiu Zhanga,b )

Preprint submitted to Ad Hoc Networks July 4, 2016

 ACCEPTED MANUSCRIPT

intersection or slippery roads [7]. Also a global positioning system device is often placed in a node of VANETs to
provide location information.
Routing plays an important role in VANETs. A good routing scheme presents reliable performance in receiving
and sending message from a source node to a destination node [8]. Clearly, to accomplish reliable message routing,
a sufficient number of nodes is required to constitute strong paths from source nodes to destination nodes. Existing
researches mainly concentrate on the routing quality in VANETs to extend network lifetime. However, little effort has
been spent to efficiently constructing routing trees. Generally, the efficiency of a distributed algorithm is assessed by
runtime and the quantity of exchanged messages among the nodes [9]. Due to the failure of sensor nodes, fast moving
vehicles, or battery exhaustion, VANETs require a routing tree being constructed as fast as possible, and runtime
becomes an important measure.

T
As above mentioned, efficient minimum spanning tree (MST) construction is very useful in routing problem of
VANETs. Moreover, clustering is a helpful technique in improving the network scalability. MST could be applied to

IP
find the ideal path from cluster head to base station in wireless sensor networks. Classical MSP algorithms in graphic
theory include Kruskal algorithm, prim algorithm, sollon algorithm and so on. All of them are greedy methods, and

CR
they could generate one and only MST. In VANETs, some nodes may escape the scope of base station or roadside
unit, and their reliability is in low level. In this case, finding a group of MST or nearly minimum spanning trees is
more applicable than classical algorithms. Swarm intelligence algorithms are inspired from collective behaviors of
natural swarms. They are good at searching a group solutions for optimization problems with no prior knowledge.

US
Artificial bee colony (ABC) is one of the commonly used swarm intelligence paradigms. In the beginning, it is
designed to handle continuous problems [10]. To efficiently deal with MST problem, this paper designs a binary
coded ABC algorithm. A novel search equation is proposed to avoid producing out of bound solutions. Moreover, the
average hitting time of the proposed algorithm for solving MST problem is discussed in this paper.
AN
The rest of the paper is organized as follows. Section 2 reports three types of VANET architecture and spanning
tree problem. Section 3 presents the proposed algorithm and its analysis. Section 4 reports the simulation results
compared with Kruskal algorithm. The average hitting time to optimal solution of spanning tree problem is also
discussed in this section. Section 5 gives the conclusion.
M

2. VANETs Architecture and Minimum Spanning Tree Problem

Ad hoc network is a new networking mode but is difficult to analyze. Gao et al. analyzed the local delay of ad
ED

hoc networks with large and small primary traffic [11]. Gorrieri et al. analyzed the performance of clustered vehicular
sensor networks, and found some insights in the trade-off between decision delay and energy efficiency [12]. Cornejo
et al. studied a reliable neighbor discovery layer for MANETs, which could be applied in regular square region or
hexagonal tiling region [13]. Han et al. investigated the relay placement and power allocation for cooperative relaying
PT

[14]. Lachowski et al. implemented distributed algorithms for building routing spanning trees in wireless sensor
networks [15]. Kponyo et al. studied a VANET cluster-on-demand MST based on prim algorithm [16]. Wang and
Liang investigated node transmission limit of hybrid wireless networks, a base station is in charge of long distance
communications of nodes [17].
CE

The architecture of VANETs can be classified to three prototypes: RVC, IVC, and HVC. In RVC, vehicles com-
municate with server through roadside access point or base station to acquire information. Due to the fast moving
property of vehicles in VANETs, each node frequently escapes from current access point and breaks into the covering
field of another access point, which is usually called handover problem. Suppose a node lies in covering fields of more
AC

than one access points, how to choose a proper access point to reach maximum transmission gain is another problem
to be solved in RVC.
In IVC, a vehicle could acquire necessary information from other vehicles when access points or base stations
are unavailable. Communication routing is the main problem in this type. IVC routing includes unicast routing,
cluster-based routing, geocast routing, and broadcast routing.
HVC is a more realistic scene in daily life. If a vehicle runs out the service scope of roadside access points and
base stations, it could connect to server through other vehicles, which are considered as routers. This is an economic
solution than deploying access points or base stations.
The MST problem has been studied for decades, researchers still attempt to design an algorithm to tackle it in
P
linear time [18]. The problem is described as follows. Given an undirected graph G = (N, E, W), where W = e∈E we
2
 ACCEPTED MANUSCRIPT

T
IP
CR
US
AN
Figure 1: Flow chart of standard ABC algorithm.

is weight function of defined on edge set E. A tree T = (N, ET , WT ) is called a spanning tree if it contains all nodes
M

P
in graph G, where WT = e∈ET we is the weight function of T . Among all spanning acyclic trees of G, the one having
the least total weight is called minimum spanning tree of G. For convenience, G is assumed to be a connected tree in
this paper.
ED

In real world applications, different MST represents different design schemes. A spanning tree having nearly
minimal total weight to MST may also be valuable to practitioners based on specific conditions. Classical MSP
algorithms could not provide several candidate spanning trees. Also it is impractical to search candidate spanning
trees by brute force method. Swarm intelligence approaches show good performance in dealing with combinatorial
PT

optimization problems. They can obtain a set of (nearly) optimal solutions in a short time. ABC is taken in this paper
to tackle the MST problem in VANETs.
CE

3. Binary Artificial Bee Colony

The colony of standard ABC consists of three kinds of honey bees. They are employed bees, onlooker bees, and
scout bees, hence the algorithm is comprised of three stages as shown in Fig. 1. They work one after another to
constitute a cycle in ABC as illustrated in Algorithm 1. ne is the number of employed bees. limit is the number of
AC

consecutive iterations that a food source fails to be improved. Generally, the number of food sources is ne , so is the
number of onlooker bees. The number of scout bees depends on limit and the algorithm’s search status. In standard
ABC algorithm, onlooker bees conduct local exploitation to refine solutions, scout bees perform global exploration to
detect promising search area, while the search by employed bees is in the middle. It can either be exploration if food
sources were scattered widely in search space, or exploitation if food sources were close to each other.
Standard ABC is targeted for continuous optimization problems. To deal with the MST problem, binary represen-
tation is utilized. Each element corresponds to an edge in graph G and takes value 0 or 1, where value 0 means the
associated edge is not included in spanning tree T , and value 1 means the opposite. Each edge e ∈ E is assigned an
integer flag so that all edges are coded into a binary string of length |E|. Thus, a tree T can be built given such kind of
binary string.
3
 ACCEPTED MANUSCRIPT

Input: f (·), D, xmin , xmax , ne , limit, f eval = 0

Output: a set of optimal solutions obtained by the algorithm
1 randomly initialize ne food sources, f eval = f eval + ne ;
2 evaluate the function values of the initialized food sources, and compute their fitness values;
3 while termination criteria are not satisfied do // main cycle
4 for i ← 1 to ne do // employed bee phase
5 employed bee i searches around food source i based on some rules and produces vi ;
6 evaluate vi , f eval = f eval + 1;
7 greedy selection between vi and xi ;

T
8 end
9 for i ← 1 to ne do // onlooker bee phase

IP
10 onlooker bee i chooses a food source based on the fitness of food sources;
11 onlooker bee i searches around the chosen source k based on some rules and produces vi ;
12 evaluate vi , f eval = f eval + 1;

CR
13 greedy selection between vi and xk ;
14 end
15 for i ← 1 to ne do // scout bee phase
16 if food source i has not been improved in the last limit iterations then
17
18
19
end
end
replace food source i by xi US 0
a scout bee flies out and randomly explores the search space to produce xi ;
0
AN
20
21 end
Algorithm 1: Procedures of the ABC algorithm
M
ED

Figure 2: Example of initializing a binary string.

PT
CE
AC

Figure 3: Example of initializing a binary string.

A simple heuristics is applied to quickly produce a spanning tree. Because a spanning tree T connects all nodes
without loop, it must contain |N| − 1 edges. Thus, a binary bit string stands for a spanning tree under the condition
that it contains |N| − 1 1s in the string. This heuristics is used in the initialization step as well as search steps in main
cycle.
In initialization, a bit string of length |E| is created with each bit set to 0. Then randomly choose |N| − 1 positions
and set those to 1. For example, given a graph G = (N, E) with |N| = 5 and |E| = 9 as shown in Fig. 2, where the
nodes and edges are numbered in order so that each bit of a solution could be decoded to an edge. Figure 3 presents

4
 ACCEPTED MANUSCRIPT

Figure 4: Tree structure represented by a binary string as in Fig. 3.

T
IP
CR
US
Figure 5: Example of two-element variation technique.

the procedure of generating an initial solution. The initialized string could be decoded to a tree having |N| − 1 edges,
AN
which is a spanning tree if all nodes are connected, or a tree with loop. Accordingly, the associated tree is shown in
Fig. 4, which is a spanning tree of graph G.
Following the above heuristics, a two-element variation technique is proposed for the search of employed bees and
onlooker bees. First, a position is randomly chosen and denoted as i1 . Second, another position i2 is also randomly
chosen depending on the value of i1 . If i1 = 0, then i2 is selected from the elements with 1s; otherwise, i2 is selected
M

from those with 0s. This technique is shown in Fig. 5. In this figure, graph G contains 5 nodes and 9 edges, hence bit
string length L = 9. It is observed that the number of elements with value 0 or 1 does not change, hence this technique
always produces a candidate string with 5 elements with value 1.
ED

If the limit flag of some food sources is satisfied, the associated food sources would be discarded, and are replaced
by new ones randomly exploration by scout bees. The procedure is similar to the initialization step. Elitism is NOT
used in scout bee phase. That is the best solution could be abandoned by honey bees, but the best solution is memorized
at the end of each cycle.
PT

The pseudo code of binary ABC (BABC) algorithm is similar to standard ABC as in Algorithm 1 except food
sources are encoded by bit strings and D = L.
To deal with the hybrid flowshop problem, a discrete ABC algorithm is designed combining differential evolution
and variable neighborhood search methods [19]. Pan et al. also presented a discrete ABC algorithm for hybrid
CE

flowship problem, which used a combination of forward decoding and backward decoding methods [20]. A bitwise
operation (e.g., XOR operator ’∧’, or operator ’|’ and operator ’&’ ) based binary ABC was proposed for continuous
function optimization [21]. ABC was also modified to optimize binary halftone patterns of images [22]. Ozturk et
al. proposed to use similarity information of solutions to improve the performance of binary ABC and verified it
AC

on dynamic clustering [23]. Instead of designing binary ABC, Kiran applied continuous ABC to handling binary
optimization problems [24].
Similar to existing works, binary representation is used in BABC. The variation search shown in Fig. 5 is different
from previous researches. Because BABC is designed to handle spanning tree problems, the number of 1 elements
during variation search has to be constant. Moreover, the generation of initial solutions is also specific to spanning
tree problems.

4. Simulation

In this section, the proposed BABC algorithm will be investigated by an MST example of RVC type networks.
5
 ACCEPTED MANUSCRIPT

T
IP
Figure 6: A VANET example with 16 nodes and 32 edges.

CR
4.1. Test Example and Simulation Setting
A graph G of 16 nodes are taken as an example in VANETs as shown in Fig. 6. This example includes 16 nodes
and 32 edges. Among the 16 nodes, one is a base station and the others are cars in a traffic way. The number of edges

US
|E| is . The weight function is Euclidean distance between two nodes. The objective function of a binary string is the
sum of weights over all edges. The MST problem is turned into a minimization model.
The parameter setting of the BABC algorithm is:
AN
(1) ne = 10;
(2) limit = 100.
The algorithm is independently executed 25 times to obtain its average performance. The soundness of this parameter
setting has already been demonstrated as in [25, 26]. It terminates when either the following conditions is met:
M

(1) the maximum number of function evaluations (MFE) is reached, where MFE = 10000;
(2) | f (x)− f (x∗ )| ≤ 10−8 , where f (x) stands for the best value found by BABC and f (x)∗ is the optimal value found
by Kruskal algorithm.
ED

Average hitting time means the average running time of BABC reaching optimal value. The following metric is
used, which is also called expected running time (ERT) [27]:
1 − p̂ s
PT

E(RT (∆ f )) = Ê(FE s ) + Ê(FE u ) , (1)

p̂ s

where RT (∆ f ) denotes the runtime of the algorithm reaching ∆ f = 10E −8 ; Ê(FE s ) means the average number of
CE

functions evaluations that those runs successfully reach ∆ f ; Ê(FE u ) is the average number of function evaluations in
the unsuccessful runs; p̂ s is the ratio of runs that reaching ∆ f over 25 runs. In this paper, Ê(FE u ) = 10000.
The simulation is conducted on a PC with 4-core 2.50GHz CPU and 4GB of memory. Thus, a fair environment
can be made for runtime comparison.
AC

4.2. Simulation Results

First, Kruskal algorithm is applied to handle the example and the resulting MST is shown in Fig. 7. Since this
method always finds MST of an undirected acyclic graph, the solution is considered as x∗ . Its objective function value
f (x∗ ) = 1949.25. The computational time of Kruskal algorithm is about 0.05 second.
Table 1 presents a summary of the results attained by the BABC algorithm over 25 runs. In the table, BABC
successfully finds the same MST solution as Kruskal algorithm in 23 out of 25 runs, which is 92%. Although in the
remaining two runs, the algorithm fails to reach the same solution, the difference is negligible in terms of the weights
of spanning trees. The spanning tree represented by the second solution in Table 1 is plotted in Fig. 8. It can be seen
that node N12 serves as a relay node for N15 and N16 in Fig. 7, while the relay node changes to N14 in Fig. 8. In
6
 ACCEPTED MANUSCRIPT

T
IP
Figure 7: Minimum spanning tree found by the Kruskal algorithm.

CR
US
AN
M

Figure 8: The second minimum spanning tree found by the BABC algorithm.
ED

case that N12 is unstable and becomes unavailable, without performing another spanning tree search, the solution as
in Fig. 8 could take over the information communication if N14 keeps stable.
The average number of function evaluations (FE) is 3219 in the 23 successful runs. Clearly, the success rate of
BABC reaching MST is 92%. The average hitting time computed by (1) is about 4088 in 25 runs, which means
PT

that the BABC algorithm could find the optimal solution costing 4088 FEs on average. The corresponding computer
running time is about 0.32 second.
Table 2 presents the results attained by the BABC algorithm in a typical run. It is observed that BABC can obtain
ten candidate spanning trees for users in one run, due to ne = 10. See from the table, several good spanning trees
CE

could be attained by executing BABC once. This is useful under the condition that spanning tree needs rebuilding
frequently yet the network topology is unchanged. BABC consumes more runtime than Kruskal algorithm, though, it
might be a good choice considering reconstructing MST repeatedly.
The BABC algorithm is also apt to deal with optimal routing from a source node to a destiny node, which is
AC

equivalent to shortest path problem. Take the base station N8 as starting node, while N1 or N16 is taken as destiny
node, respectively. For verifying the performance of the proposed algorithm, the optimal path from N8 to N1 or N16
are manually determined as shown in the third column of Table 3. ERT of the algorithm can then be computed. The
results are given in Table 3.

Table 1: Summary of spanning trees found by the BABC algorithm in 25 runs

ST Weight Edge series Runs
1 1949.25 E1 E4 E7 E8 E9 E10 E14 E17 E18 E20 E25 E26 E27 E28 E32 23
2 1949.35 E1 E4 E7 E8 E9 E10 E14 E17 E18 E20 E25 E26 E28 E30 E32 2

7
 ACCEPTED MANUSCRIPT

Table 2: A set of spanning trees found by the BABC algorithm in a typical run
ST Weight Edge series
1 1949.25 E1 E4 E7 E8 E9 E10 E14 E17 E18 E20 E25 E26 E27 E28 E32
2 1950.11 E1 E4 E6 E7 E8 E10 E14 E17 E18 E20 E25 E26 E27 E28 E32
3 2035.04 E1 E4 E5 E7 E10 E12 E14 E18 E20 E22 E25 E26 E27 E28 E32
4 2078.48 E1 E3 E5 E8 E9 E10 E14 E17 E18 E20 E25 E26 E27 E28 E32
5 1958.72 E1 E4 E5 E7 E10 E12 E14 E17 E18 E20 E25 E26 E27 E28 E32
6 2154.08 E1 E2 E5 E7 E9 E10 E15 E17 E18 E20 E24 E26 E28 E31 E32
7 2554.57 E2 E3 E5 E7 E10 E11 E12 E18 E20 E22 E24 E26 E27 E30 E31
8 2423.31 E1 E3 E6 E7 E8 E11 E15 E17 E18 E20 E23 E25 E27 E30 E31
9 2297.14 E1 E3 E4 E5 E9 E10 E14 E17 E21 E22 E25 E26 E27 E30 E32

T
10 2220.11 E2 E4 E6 E7 E8 E10 E13 E17 E18 E20 E24 E26 E27 E28 E32

IP
Table 3: Shortest path test of the BABC algorithm in 25 runs
Source Destiny Shortest path Runs ERT
N8 N1 N8 N5 N2 N1 25 251.80

CR
N8 N16 N8 N13 N14 N16 25 409.24

It can be seen from Table 3 that the success rates of BABC finding optimal path from N8 to N1 or to N16 are 100%.

US
The ERT for path N8 to N1 is about 251.8 over 25 runs, which means that the BABC algorithm could find shortest
path costing about 252 FEs on average. The ERT for path N8 to N16 is about 409.24 over 25 runs, which means that
the algorithm could find shortest path costing about 410 FEs on average. Although the proposed algorithm can be
applied to deal with optimal routing in VANETs, it has a large variance based on the ERT values.
AN
5. Conclusion

Efficient and reliable communications in VANETs requires constructing strong paths among vehicles. Due to the
M

failure of sensor nodes, fast moving vehicles, or battery exhaustion, some nodes may become unavailable of ad hoc
network. New paths has to be built to assure information communication in ad hoc network. Path construction can be
modeled as minimum spanning tree (MST) problem in graphic theory. Classical algorithms in graphic theory could
find MST but cannot provide alternatives for usage. Note that the artificial bee colony (ABC) paradigm is able to
ED

accomplish this target.

This paper proposes a ABC algorithm with binary representation (BABC). VANETs are represented by undirected
acyclic graphs. Edges of a graph are encoded to a binary string in BABC. To efficiently exploring all possible spanning
tree combinations, a two-element variation technique is designed to keep the number of edges constant in a binary
PT

string.
Tested on an example of roadside-to-vehicle communication networks, the BABC algorithm finds the optimal
spanning tree as using Kruskal algorithm. Moreover, BABC could also produce candidate suboptimal spanning trees.
These trees are useful for path construction when some nodes in ad hoc network becomes unavailable. Although the
CE

BABC algorithm costs more computational time than Kruskal algorithm, it would be a useful tool when routing paths
have to be build repeated in a short time. Furthermore, the algorithm can also be applied to deal with shortest path
problem in VANETs.
The average hitting time of the BABC algorithm could be reduced by designing more effective search equations
AC

or by identifying search patterns in binary strings. For networks with a large number of nodes, it may be better to add
prior knowledge about network structure to enhance the efficiency of BABC. These issues will be further studied in
the future.

Acknowledgements

This research was supported in part by the Applied Basic Research Program of Tianjin (15JCYBJC51500, 15JCY-
BJC52300), and the Tianjin Thousand Youth Talents Plan Project of Tianjin Normal University (ZX110023).

8
 ACCEPTED MANUSCRIPT

References
[1] X. Wang, Q. Liang, On the throughput capacity and performance analysis of hybrid wireless networks over fading channels, IEEE Trans on
Wireless Communications 12 (6) (2013) 2930–2940.
[2] Q. Liang, X. Cheng, S. Huang, D. Chen, Opportunistic sensing in wireless sensor networks: theory and applications, IEEE Transactions on
Computers 63 (8) (2014) 2002–2010.
[3] Q. Liang, Radar sensor wireless channel modeling in foliage environment: UWB versus narrowband, IEEE Sensors Journal 11 (6) (2011)
1448–1457.
[4] Q. Liang, X. Cheng, S. Samn, NEW: Network-enabled electronic warfare for target recognition, IEEE Transactions on Aerospace and
Electronic Systems 46 (2) (2010) 558–568.
[5] Q. Liang, Situation understanding based on heterogeneous sensor networks and human-inspired favor weak fuzzy logic system, IEEE Systems
Journal 5 (2) (2011) 156–163.

T
[6] Q. Liang, X. Cheng, KUPS: Knowledge-based ubiquitous and persistent sensor networks for threat assessment, IEEE Transactions on
Aerospace and Electronic Systems 44 (3) (2008) 1060–1069.

IP
[7] S. Bitam, A. Mellouk, S. Zeadally, HyBR: a hybrid bio-inspired bee swarm routing protocol for safety applications in vehicular ad hoc
networks (VANETs), Journal of Systems Architecture 59 (10) (2013) 953–967.
[8] R. Lachowski, M. E. Pellenz, M. C. Penna, E. Jamhour, R. D. Souza, An efficient distributed algorithm for constructing spanning trees in
wireless sensor networks, Sensors 15 (1) (2015) 1518–1536.

CR
[9] M. Khan, G. Pandurangan, V. S. A. Kumar, Distributed algorithms for constructing approximate minimum spanning trees in wireless sensor
networks, IEEE Trans. Parallel Distrib. Syst. 20 (2009) 124–139.
[10] D. Karaboga, B. Basturk, A powerful and efficient algorithm for numerical function optimization: artificial bee colony (ABC) algorithms,
Journal of Global Optimization 39 (3) (2007) 459–471.
[11] J. Gao, C. Yin, Analysis of the local delay with slotted-aloha based cognitive radio ad hoc networks, EURASIP Journal on Wireless Commu-
nications and Networking 2015 (1) (2015) 1–6.

Networks 36, Part 2 (2016) 450–464. US

[12] A. Gorrieri, M. Martalo, S. Busanelli, G. Ferrari, Clustering and sensing with decentralized detection in vehicular ad hoc networks, Ad Hoc

[13] A. Cornejo, S. Viqar, J. L. Welch, Reliable neighbor discovery for mobile ad hoc networks, Ad Hoc Networks 12 (2014) 259–277.
[14] L. Han, J. Mu, W. Wang, B. Zhang, Optimization of relay placement and power allocation for decode-and-forward cooperative relaying over
AN
correlated shadowed fading channels, EURASIP Journal on Wireless Communications and Networking 2014 (1) (2014) 1–7.
[15] R. Lachowski, M. E. Pellenz, M. C. Penna, E. Jamhour, R. D. Souza, An efficient distributed algorithm for constructing spanning trees in
wireless sensor networks., Sensors 15 (1) (2015) 1518–1536.
[16] J. J. Kponyo, Y. Kuang, E. Zhang, K. Domenic, VANET cluster-on-demand minimum spanning tree (MST) prim clustering algorithm, in:
International Conference on Computational Problem-solving (ICCP), 2013, Jiuzhai, China, 2013, pp. 101–104.
M

[17] X. Wang, Q. Liang, On the throughput capacity and performance analysis of hybrid wireless networks over fading channels, IEEE Transac-
tions on Wireless Communications 12 (6) 2930–2940.
[18] S. Pettie, V. Ramachandran, An optimal minimum spanning tree algorithm, Journal of the ACM 49 (1) (2002) 16–34.
[19] Z. Cui, X. Gu, An improved discrete artificial bee colony algorithm to minimize the make span on hybrid flow shop problems, Neurocomput-
ing 148 (2015) 248–259.
ED

[20] Q.-K. Pan, L. Wang, J.-Q. Li, J.-H. Duan, A novel discrete artificial bee colony algorithm for the hybrid flowshop scheduling problem with
makespan minimisation, Omega 45 (2014) 42–56.
[21] D. Jia, X. Duan, M. K. Khan, Binary artificial bee colony optimization using bitwise operation, Computers & Industrial Engineering 76
(2014) 360–365.
[22] A. Chatterjee, B. Tudu, K. C. Paul, Binary grayscale halftone pattern generation using binary artificial bee colony (bABC), Signal Image and
PT

Video Processing 7 (6) (2013) 1195–1209.

[23] C. Ozturk, E. Hancer, D. Karaboga, Dynamic clustering with improved binary artificial bee colony algorithm, Applied Soft Computing 28
(2015) 69–80.
[24] M. S. Kiran, The continuous artificial bee colony algorithm for binary optimization, Applied Soft Computing 33 (2015) 15–23.
[25] M. A. Contreras-Cruz, V. Ayala-Ramirez, U. H. Hernandez-Belmonte, Mobile robot path planning using artificial bee colony and evolutionary
CE

programming, Applied Soft Computing 30 (2015) 319–328.

[26] X. Zhang, X. Zhang, S. L. Ho, W. N. Fu, A modification of artificial bee colony algorithm applied to loudspeaker design problem, IEEE
Transactions on Magnetics 50.
[27] N. Hansen, A. Auger, R. Ros, S. Finck, P. Pošı́k, Comparing results of 31 algorithms from the black-box optimization benchmarking bbob-
AC

2009, in: Proceedings of the 12th Annual Conference Companion on Genetic and Evolutionary Computation, GECCO ’10, ACM, New York,
NY, USA, 2010, pp. 1689–1696.

9
 ACCEPTED MANUSCRIPT

Author-Biography
Xin Zhang received the B. Sc., M. Sc., and Ph.D. degrees from from Ludong
University, Shandong University of Science and Technology, City University of Hong
Kong, in 2006, 2009, and 2013, respectively. He is currently a Lecturer in the College
of Electronic and Communication Engineering, Tianjin Normal University.
His main research interests are swarm intelligence, communication network
optimization, evolutionary computation, and machine learning. He has published
more than 15 technical papers on these subjects, including more than eight
international journal papers.

T
IP
Xiu Zhang received the B. Eng. and M. Eng. degrees in biomedical engineering from

CR
the Hebei University of Technology, Tianjin, China, in 2006 and 2009, respectively.
Her research during master period concerns the analysis of the electroencephalograph
signal when the magnetic signal is used to stimulate the acupoints in human subjects.

US
She received the Ph.D. degree and completed the postdoctoral research work in
electrical engineering from The Hong Kong Polytechnic University in 2012 and 2015,
respectively.
She is now a lecturer in Tianjin Normal University. She has published about 20 papers
AN
in the IEEE TRANSACTIONS. Her research interests mainly focus on numerical
methods of electromagnetic field computation, novel wireless energy transfer
systems, and wireless network optimization.
M
ED
PT
CE
AC
 ACCEPTED MANUSCRIPT

Author-Photo

T
IP
Xin Zhang

CR
US
AN

Xiu Zhang
M
ED
PT
CE
AC