Upload
jing-jing
View
213
Download
0
Embed Size (px)
Citation preview
An Ant-colony Routing Algorithm for Wireless Sensor Network
Guangcai Cui1, a, Shanshan Wang1, b and Jingjing Fang1, c 1Changchun University of Science and Technology, Jilin, Changchun, 130022, China
Keywords: Wireless Sensor Network (WSN), Ant-colony Algorithm (ACO), The Optimal Routing, Energy Consumption, Negative Feedback Mechanism, Time Delay.
Abstract. According to real-time and limited energy of the wireless sensor network(WSN), this paper
proposed an ant-colony algorithm (ACO) for optimal routing. The algorithm limited the search space
to next node based on search angle and designed directional pheromones to guide ants to the
destination node. Using negative feedback mechanism encouraged later ants to choose the optimal
path. When ants are timeout with limited life cycle, go back along the way and reduce the pheromone.
Probability-transfer function contained the factors of distance, energy, pheromones and search angle.
Compared with other ACOs, the results show that it can balance the energy consumption and improve
the routing in aspects of energy, dead nodes, short path and time delay.
Introduction
WSN is composed of a set of limited-energy sensor nodes, and is multiple hops self-organizing
network by wireless connect. With the development of the Internet of things, WSN as key technology
has a wider application prospect. But in environment without supervision and the nodes energy very
limited, providing how to balance the energy consumption and create a relatively stable work
environment for WSN, is the key issues related to truly practical WSN.
In WSN, [1] designed probability-transfer function according to the energy consumption of nodes
communication on paths and remaining energy status. It made the network energy consumption more
balanced, but the search space is large and convergence speed is slow. [2] created ant-colony routing
algorithm by remaining energy of adjacent nodes and the number of routing hops. To balance the
energy consumption is the network guarantee, but the algorithm affects the search speed of ants based
on the typical design of non-directional pheromones. [3] (MP-ACRA) studied routing algorithm for
WSN from the aspect of the multi-path routing and proposed multi-path routing algorithm without
intersecting edges. This algorithm can prolong the network lifetime, but cannot guarantee the global
optimal path. [4] (ES-ACRA) improved pheromones update process and considered the energy
distribution of all nodes in the routing. In the whole network energy balance is improved, but on the
optimal path it has no guarantee. [5] created ant colony routing algorithm by the total energy
consumption of messages in the routing. It can minimize the network energy consumption, but there
are no feedback messages of dead ants to improve the performance of the network.
For the above shortages this paper proposed an improved an-colony routing algorithm from the
aspects of communication distance, directional pheromones, energy, limited life cycle and search
angle. This algorithm can narrow the search scope, balance the nodes energy consumption, search the
optimal path of energy conservation fast and prolong the lifetime of WSN.
Typical ACO
Idea of ACO. Ants have features of self-organization, self-adaption, mutual communication and
collaboration. ACO simulates their intelligent behavior and can solve many complex problems. ACO
[6] is proposed as a new type of simulated evolutionary algorithm in the early 1990s, is also a very
important method of cluster analysis. It has strong extensibility and robustness and can adapt to the
dynamic environment, so it is suitable for the design of dynamic WSN. ACO can simulate the real
process of a large number of ants find food together. Ants select paths by random and leave
Applied Mechanics and Materials Vols. 462-463 (2014) pp 112-117Online available since 2013/Nov/15 at www.scientific.net© (2014) Trans Tech Publications, Switzerlanddoi:10.4028/www.scientific.net/AMM.462-463.112
All rights reserved. No part of contents of this paper may be reproduced or transmitted in any form or by any means without the written permission of TTP,www.ttp.net. (ID: 130.207.50.37, Georgia Tech Library, Atlanta, USA-12/11/14,15:01:15)
pheromones on the search path. Then ants make a path decision by the pheromones of olfactory
perception. The more quantity of pheromones on path, the greater probability the path is selected. So
it realizes the clustering process of self-organization.
Advantages of ACO for WSN. This paper analyzed the features of WSN and ACO. WSN is a
self-organizing network and need to rely on each node in the network of cooperation to complete a
task. A single disabled node will not affect the other nodes and the whole network. In the process of
looking for routing, ACO can find the shortest path from source to destination and can save energy in
maximum. At the same time it also can improve the reliability of data transmission and the
adaptability to the change of network topology. Routing choice depends on the dynamic amount of
pheromones and is a directed measurement for the situation of real-time network. Therefore, it can
improve the real-time performance of network.
Improved Ant-colony Routing Algorithm for WSN
Strategy. The lack of typical ACO in: considering node energy consumption is not enough, balancing
the network energy consumption is not well; search speed needs to be improved; pheromones update
strategy needs to be modified to improve the convergence speed of algorithm; When the algorithm
run after a period of time, the nodes energy easily work out on the shortest path. So this paper
proposed the improved strategy and can effectively solve the above problems.
Search Angle. If ants select the node with the greatest transfer probability, the node maybe not the
nearest distance to sink in the [1]. Thus the total path length of data transfer and the total energy
consumption increases, at the same time, it does not conform to the principle of the shortest path in
routing algorithm design.
Search angle is the intersection angle θ composed of the next node, this node and the sink. The
smaller θ, the shorter the search path and the less energy consume. It also limits the search space of
next node and accelerates the search speed. As a result, this paper designed the formula (1+cosθ)λ ,
when θ is smaller, the function value is greater. Directional Pheromones. The traditional pheromones on path are no direction. As shown in Fig.1,
the pheromones amount the ants on node A perceived on path AB is equal to the pheromones amount
ants on node B perceived on path BA. If there is an ant from node C to node B, it will choose next
node A and reach the node A according to the probability-transfer function. It does not choose sink D,
largely because path BA and path BD have the same amount of pheromones. Therefore, the ant may
deviate from the sink and the more walk the more far.
Directional pheromones refer to the pheromones is not necessarily equal on two directions of the
same path. As shown in Fig.1. When the ants reach the sink and return through DB, BA and AS, but
only release pheromones in the direction of BD, AB and SA. In this way, when the later ants reach
node B, they would be not affected by pheromones on path BA.
Pheromones Update. Ants release pheromones after they reached the sink and returned in [3].
pheromones volatilization can not timely guide later ants to choose another path, so it reduces the
convergence speed. Moreover, if the path has more pheromones, the node on the path may be has little
residual energy. The result easily falls into local optimal solution.
This paper divides ants into forward ants and backward ants. Forward ants pass through nodes and
consume pheromones of paths. Its formula is (1). Backward ants pass through nodes and release
pheromones of paths. Its formula is (2).
(1)
(2)
Applied Mechanics and Materials Vols. 462-463 113
This formula shows the pheromones when backward ants pass through path ij and release. Among
them, C is a constant, AllE is the total remaining energy of all the nodes on the paths ants passed,
DiffuseE refers to the total energy consumption of the path. Therefore, the more AllE, the less
DiffuseE on the path and the more pheromones ants released.
Probability-transfer Function. In conclusion, probability-transfer function is affected by energy,
nodes distance, pheromones, search angle and other factors. So we design the formula (3).
(3)
(4)
In formula (3), ej(t) means the remaining energy when ants reach node j and consume pheromones
at t time. When residual energy of nodes is less than the minimum Emin, the nodes is disabled and
remove its information from all of its adjacent nodes. ηij is the reciprocal of the distance (dij) between
nodes. Routei table records all adjacent nodes of node i within the transmission radius (Dr). Compute
each probability node i can reach its adjacent nodes, choose the adjacent node with the greatest
probability as next node and store in the table BestRoute.The function can balance the influence of
various factors and maintain optimization of network performance.
Negative Feedback Mechanism. When [4] determines the optimal paths in this stage, the WSN
data will transmit to sink along the optimal paths. Data transmission will consume node energy along
the paths. If a large amount of data transmits along the optimal paths, it will accelerate energy
consumption of nodes on the optimal paths and cause energy imbalance of the entire network.
This paper used negative feedback mechanism of ants [7], exist behavior reduce later behavior, to
solve the contradiction between positive feedback and negative feedback. When a message arrives at
node i, we assume path ij is one of the optimal paths, there is formula (5).
(5)
In formula (5), diffusion means reducing pheromones. The function is to reduce for next path and
to strengthen pheromones for paths of other adjacent nodes. It encourages later ants to choose new
optimal paths and balances data flow. So the strategy solves the problem energy consumes fast on the
shortest paths and improves performance of the algorithm.
Life Cycle of Ants. If the ant does not reach sink and its time is over, then delete the message in
[5]. It does not make better use of the ants for feedback to warn later ants not to pass the way which
cannot arrive at sink.
To prevent the ant message time is too long, when the ant moves to next node, its life cycle
decreases one until zero. If the ant does not reach sink and the life cycle is timeout, then returns along
the same paths and reduces the pheromones, at the same time adds the pheromones for paths of other
adjacent nodes. This update operation is able to solve the above problem and encourages later ants to
open new optimal paths and to rapidly arrive at sink.
114 Progress in Mechatronics and Information Technology
Algorithm Description
Assumed that all nodes in WSN are homogeneous, have the function of data fusion, initialize the
same limited energy, and can get their geographic coordinates by GPS devices. Each node maintains a
table, the coordinates of its adjacent nodes and can periodically exchange a small amount of data to
refresh the information with adjacent nodes in the table. When the energy of node is lower than a
certain value, the distance between this node and its adjacent nodes is set with the maximum number. Define the structure of ants in Table 1:
The algorithm sents N ants as a cycle. In this cycle, ants are numbered from 0 at start. After ants
return to the source, they will be record the routing paths from source to sink until ant N-1 returns.
Then it judges the optimal paths from the N paths and store in the bestRoute array. The next cycle of
ants are numbered from 0 again. At the end of each cycle, we can transmit data. Algorithm flow chart
is shown in Fig.2. Initialization work includes the initial value of every parameter, the calculation of
distance between each node, and the establish of a neighbor information table. Then we add all nodes
within Dr Distance into the current node information table and calculate the cosθ. When a node is
disabled in the network, just delete it from the tables of neighbor nodes.
Table2. Network life cycle
Analysis Results
In order to verify the algorithm validity, we simulated MP-ACRA in [3], ES-ACRA in [4] and
IACRA in this paper. In the experiments, we used the same initial conditions and data for the above
three algorithms and to avoid accident, each experiment does 20 times and gets the means. We choose
Table1. The structure of ants
Source Destination Route table Energy Flag for forward
ants
The length
of paths
Life cycle
Fig.2. Algorithm flow chart
Fig.1. Release the directional pheromones
Applied Mechanics and Materials Vols. 462-463 115
the eil51 instance from TSP repository as experiment data. The size of transmission data is 64 Byte
every time. 8 ants all move from the source at start. Set the initial values of parameters as follows: α=
0.1, β= 2, γ= 1, λ= 1, ρ= 0.05, δ= 0.0005, Dr = 20 m, e = 5J. Parameters of the above three algorithms
must be the same, in order to make sure the results with the same convergence and complexity.
Life Cycle of Network. This paper defines the network life cycle is from run to the p% node death.
We record p in 10, 20, 50 and 80. When p is 10, it means nodes begin to appear a small amount of
death for a period of time and the network performance is very good. Compared with lifetime of the
three algorithms, the longest one means network energy is balanced. When p is 80, it means almost all
of nodes died and the network cannot work for data transmission. The longer the algorithm runs, the
longer network lifetime is. The table 2 is given network life cycle for MP-ACRA, ES-ACRA and
IACRA in the node number of 50 and 100.
By Table 2, we can see that life cycle of IACRA slightly decreases with the increasing number of
nodes and is still relatively stable on the whole. ES-ACRA fell sharply with the increase of network
scale. MP-ACRA may lead to instability of network due to establish of multiple paths when nodes are
few. For a large number of nodes, due to the large network space and energy consumption of nodes
communication, MP-ACRA will lead to reduce the life cycle further. So we can predict life cycle of
MP-ACRA will be sharply reduced if we enlarge the network scale further.
Energy Consumption. Network lifetime is proportional to residual energy of nodes as we known.
The more total rest energy of all nodes in the network at the same time the longer network lifetime.
We can see from Fig.3, when the nodes number is 5, energy consumption is almost the same for
routing build in the three algorithms. But energy consumption is the largest in MP-ACRA and the
larger in ES-ACRA as second with the increase of network scale. So energy consumption in IACRA
is smallest and reduces about 20% than ES-ACRA. IACRA has a long network lifetime mainly
depended on the minimal energy consumption of the routing establish with few ants. So it obtains the
destination of WSN protocoland it can reduce energy consumption, increase energy efficiency and
maximize the network lifetime as possible.
Fig.3. Energy consumption
Fig.4. The number of disabled nodes
Numbers of Disabled Nodes. The number of disabled nodes relates to energy consumption in the
network. It can maintain the network lifetime is longer if the number of disabled nodes is less and
disabled nodes occur later.The simulation is shown in Fig.4. Compared with the three algorithms,
IACRA has the longest lifetime and best performance of entire network. ES-ACRA considers energy
consumption insufficiently and leads to inequality of energy consumption and to kill nodes rapidly.
Under the conditions of the same initial energy, the lifetime in MP-ACRA is longer than ES-ACRA
within 25 nodes, but its nodes die rapidly later and the rest nodes are less. So MP-ACRA cannot show
its superiority in practical application. The Shortest Paths. The shorter the total length of the shortest paths, the less network energy
consumption. This paper tests the experiment by 20 times for the three algorithms respectively. The
result is shown in Fig.5. The shortest paths in IACRA is shorter and fluctuation of the final result
116 Progress in Mechatronics and Information Technology
graph is less. It means the algorithm performance is more stable than others. The simulation shows
IACRA can work out a better global optimal solution to solve the shortest paths problem with strong
robustness in WSN.
Average delay. Shorter average delay refers to the faster speed of algorithm convergence and the
less energy consumption. This paper tests experiments for the three algorithms in the different node
number and records their convergence time. The simulation is shown in Fig.6. In the network scale 40,
the average dalay for three algorithms is almost the same, but with the increase of network scale,
average delay in IACRA is lower than the other two algorithms. The main reason is that the optimal
paths in IACRA decreases the number of jump, thus it can reduce the average delay.
Conclusions
This paper applies the ACO to WSN with features of WSN and ACO and analyzes communication
distance, energy, pheromones, search angle, speed and other factors. Then it establishs
probability-transfer function and test experiments in different aspects. In results, it can balance the
node energy consumption effectively, reduce the disabled nodes, increase the convergence speed,
improve the performance of WSN and extend the lifetime of the network. Such algorithm has a high
application value and will be considered in future works on this area.
References
[1] Xirong Bao, Shi Zhang and Dingyu Xue: Research and Simulation on Genetic Ant Colony
Routing in Wireless Sensor Network. Proceedings of fourth International Conference on Wireless
Communications, Networking and Mobile Computing. Dalian, China, Sep. 19-21, 2008, pp. 32-27.
[2] S.D. Muruganathan, F.C.D. Ma and R.I. Bhasin: A Centralized Energy-efficient Routing Protocol
for Wireless Sensor Networks. IEEE Radio Communication Magazine, Vol.43, N°3, pp. 13-25, 2005.
[3] F. Li, K. Wu and A. Lippman: Energy-efficient Cooperative Routing in Multi-hop Wireless Ad
Hoc Networks. In Performance, Computing and Communications Conference. IPCCC 25th
IEEE
International, Apr. 10-12, 2006, pp. 124-136.
[4] A. Salehpour, B. Mirmobin and K. Afzali: An Energy Efficient Routing Protocol for
Cluster-based Wireless Sensor Networks Using Ant Colony Optimization. Chinese Journal of
Sensors and Actuators, Shengyang, China, Jun. 17-19, 2008, pp. 4716-4721.
[5] T. Camilo, C. Carreto and J.S. Silva: An Energy-efficient Ant-based Routing Algorithm for
Wireless Sensor Networks. Lecture Notes in Computer Science, Apr. 23-29. 2006, pp. 325-329.
[6] Hadim Salem and Nader Mohamed: Middleware Challenges and Approaches for Wireless Sensor
Networks. IEEED distributed Systems Online, Vol. 29, pp. 68-87, 2006.
[7] A.K. Kkaya and M.A. Younis: Survey on Routing Protocols for Wireless Sensor Networks. Ad
Hoc Networks, Vol. 23, N°3, pp. 305-326, 2005.
Fig.5.The shortest paths
Fig.6.Average delay
Applied Mechanics and Materials Vols. 462-463 117
Progress in Mechatronics and Information Technology 10.4028/www.scientific.net/AMM.462-463 An Ant-Colony Routing Algorithm for Wireless Sensor Network 10.4028/www.scientific.net/AMM.462-463.112