Beruflich Dokumente
Kultur Dokumente
j k
,
1,
0
1 ,1
,
S j R k
s is covered by z
X j N k N
otherwise
1
Z
N
i
i
i
Minimi e c Y z
=
(1)
Subject to:
,
1
1 2 ,
R
N
j R k
k
X j j N
=
(2)
0
1
0
Z
N
i
i
Y C
=
=
(3)
Assume ,1
Z i
c N c i = , then the objective is transformed
into minimizing the number of RNs.
IV. OPTIMIZATION STRATEGY DESIGN
A. Overview of GASA
GA has global characteristics and unique advantages in
solving NP-hard problem [13-14]. However, the standard GA
has poor fine-tuning ability which resulting in falling into local
optimum and premature convergence easily [15-16].
SA is a random search algorithm of the simulation of the
physical annealing process and it has strong local search ability
[17-18]. But with SA to achieve excellent results is not effi-
cient [19].
GASA combines the advantages of GA and SA, and it is
very effective in solving large
scale combination optimization problems [20]. GA finds gen-
erally good global solutions, SA searches local optimization of
population generated by GA. At higher temperature, GASA
shows sudden jumps of strong probability; at low temperature,
it evolves to be chemotactic local search algorithm. The proce-
dure of GASA proposed is shown in Alg. 1.
Algorithm 1: Procedure of GASA
Step 1: Initialize population size( popSize ), crossover rate
(
c
P ), mutation rate (
m
P ), initial temperature(
0
T ), gra-
dient of cooling( ), count of iterations of Metropolis
algorithm in one temperature phase( L ), the maximum
iterations of temperature cooling ( K ), elite
count( eliteCount );
Step 2: 0 k ,
0
T T , generate the initial population
( ) 0 POP ;
Step 3: Calculate the fitness of ( ) 0 POP ;
Step 4: At the current temperature T , repeat Step 5 while the
number of iterations is less than L ;
Step 5: Apply selection first, then apply crossover and muta-
tion with probability
c
P and
m
P on ( ) POP k , calcu-
late the fitness
'
i
f of new candidate individuals, set
'
, k i i i
f f f = > ,if
,
0 <
k i
f , the accept the individual
as the new one; otherwise, accept it with probabili-
ty
( )
,
/
k i
exp f T ;
Step 6: T T , 1 k k + , if k K < , go to Step 4, otherwise
terminate and output the optimal solution.
B. Encoding
In the case of RNs placement problem, an individual con-
tains the information on the current location of RNs in the grid
area. The chromosome can be represented by
{ }
1 2
, , 1 , 1 , , 0
Z
Z N i
Y Y Y Y i N =
X . If the candidate
location i is selected to place RN, the corresponding
i
Y should
be set to 1 regardless of its current value.
C. Objective function
Set
( )
1
Z
N
i
i
f Y
=
=
X (4)
and (4) subject to (2)(3)
In order to use GASA of single objective, a penalty func-
tion is needed to transform the constrained optimization prob-
lem into an unconstrained optimization problem.
In this paper, we get the single objective function with a
penalty function as following:
( ) ( )
, 0
1 1 1
, 0 max 2
R S Z
N N N
j k i
j k i
M C F f X Y
= = =
= + +
X X (5)
M is the penalty function factor. This paper does not allow
the existence of solutions which does not meet the above-
mentioned constraint conditions. Therefore, M can be arbitra-
rily large integer.
D. Selection
In this paper, selection procedure adopts the roulette wheel
approach which is one of the fitness-proportional selections.
The elitist way was combined with this approach in order to
preserve the outstanding individuals in the next population.
Parent A
Parent B
Child A Child B
Rect 1 Rect 1
SN
RN
BS
Rect 2 Rect 2
Figure 1. Example of crossover operation
E. Crossover
In this paper, the crossover operator should thus take into
account the specifics of application scenarios. Similar to [21],
the crossover operator that we use is outlined in Alg. 2. Take
Fig. 1 for example. Obviously, Rect 1 moved to Chid B has
improved the 2-connected connectivity and reduces the number
of RNs with respect to that of Parent B and Rect 2 moved to
Child A has improved the 2-connected connectivity with re-
spect to that of Parent A.
Algorithm 2: Procedure of crossover operation
Step 1: Duplicate Parent A and Parent B and saved as Child A
and Child B respectively
Step 2: Select rectangles Rect 1 and Rect 2 randomly from
parents and children respectively
Step 3: Rect 1 of Parent A and Parent B are transferred into
Rect2 of Child B and Child A respectively
F. Mutation
Similar to [22], our mutation operator is composed of four
types of mutation operations: SingleMutate, RectMutate,
SmallMutate and SmallRectMutate. In this paper, the Probabil-
ities are set to 0.1, 0.3, 0.1, 0.5.
V. EXPERIMENT
In this section, we evaluate the performance of our algo-
rithm via simulations. In all simulation scenarios, we use the
number of RNs as the performance metric because our objec-
tive is to minimize the installation cost.
The SNs were randomly distributed in a
2
200 200m
square area. The communication range of each SN
S
l is 50m
0 20 40 60 80 100 120 140 160 180 200
0
20
40
60
80
100
120
140
160
180
200
0 20 40 60 80 100 120 140 160 180 200
0
20
40
60
80
100
120
140
160
180
200
(a) Initial iteration (b) 20th iteration
0 20 40 60 80 100 120 140 160 180 200
0
20
40
60
80
100
120
140
160
180
200
0 20 40 60 80 100 120 140 160 180 200
0
20
40
60
80
100
120
140
160
180
200
(c) 40th iteration (d) 200th iteration
Figure 2. Samples of best individual during the evolution process
and the communication range of each
R
l is 50m too. Initialize
parameters: 100 = popSize
0
50 T = 0.8 = 20 L =
10 K = , 0.6 =
c
P , 0.4
m
P = , 2 elitesCount = , the total itera-
tions of GA procedure, i.e. the iterations of the inner loops with
respect to GASA, is 200 L K = .
In what follows, the square area is divided into 20 20
cells of each side 10 w m = , i.e. 21 21
Z
N = . The rectangles
in the crossover and mutation operation are squares with side
randomly generated from ( ) ,15 w w and ( ) , 5 w w respectively.
The step of SmallMutate and SmallRectMutate in the mutation
operation is 5w.The initial population of GASA is randomly
generated.
As shown in Fig. 2 where the blue plus signs denote SNs,
the green filled circles denote RNs and the green filled square
denotes the BS, we shown four samples of best individuals
found at different steps during the evolution process of one run.
It is noted that with the increase of GA iterations (the iterations
of the inner loops with respect to GASA), the search tends to-
wards the global optimal direction and the number of RNs de-
creases. At last, a feasible solution and the network topology
shown in Fig. 2(d). In Fig. 2(d), the dashed lines present the
links between SNs and RNs and the sold lines denote the links
between RNs or RNs and BS. Fig. 3 shows the process of the
number of RNs varying with growing iterations. As shown in
Fig. 3, the number of RNs converges quickly in the early stage
of the process.
We also have experienced how the number of RNs varies
with the distribution of SNs and compared the performance of
the proposed GASA with GA and ILP [10] using CPLEX [23].
All of the following performance results are averaged over 15
runs of the same initializations. The result of comparison is
shown in Fig. 4.As shown in Fig. 4, the number of RNs in-
creases with the increasing of the number of SNs for small
number of SNs; But when the number of SNs reaches a certain
number(>50), the kind of increase becomes not substantial. It is
0 20 40 60 80 100 120 140 160 180 200
15
20
25
30
35
40
45
N
o
.
o
f
r
e
l
a
y
n
o
d
e
s
No. of iterations
Figure 3. No. of relay nodes versus No. of iterations
0 20 40 60 80 100 120 140 160
10
12
14
16
18
20
N
o
.
o
f
r
e
l
a
y
n
o
d
e
s
No. of sensor nodes
ILP
GA
GASA
Figure 4. No. of relay nodes versus No. of sensor nodes
because that when the network is dense enough, there is no
necessary to add additional RNs to meet 2CDCP. Comparing
GASA with GA, the number of RNs derived from GASA is
much smaller. Comparing GASA with ILP, the number of RNs
derived from GASA is far less than that of ILP when the densi-
ty of SNs is small; with the density of SNs increases the advan-
tage becomes not substantial.
VI. CONCLUSION
In order to deploy relay nodes with minimum cost for in-
dustrial fault-tolerant hierarchical wireless sensor networks, we
proposed a genetic simulated annealing hybrid algorithm to
optimize the relay node deployment. In such a network, instal-
lation cost depends on the number of RNs. The resilience to
single link failures is considered and the proposed method en-
sures that each SN can be covered by at least two RNs, and the
network RNs is 2-connected. We have evaluated the perfor-
mance of our approach by comparing it with GA and ILP. The
results have shown that the proposed GASA quickly converges
to the feasible solution within rational time and GASA leads to
improvements compared with GA and ILP though GASA con-
sumes little more time than GA and ILP. In order to improve
the performance of industrial wireless networks, the propaga-
tion characteristics of the wireless signal (such as signal inter-
ference, capacity, etc.) should be taken into account, which will
be discussed in the further study.
REFERENCES
[1] J. R. Moyne and D. M. Tilbury, "The Emergence of Industrial Control
Networks for Manufacturing Control, Diagnostics, and Safety Data,"
Proceedings of the IEEE, vol.95, pp. 29-47, 2007.
[2] A. Willig, K. Matheus and A. Wolisz, "Wireless Technology in
Industrial Networks," Proceedings of the IEEE, vol.93, pp. 1130-1151,
2005.
[3] V. C. Gungor and G. P. Hancke, "Industrial Wireless Sensor Networks:
Challenges, Design Principles, and Technical Approaches," Industrial
Electronics, IEEE Transactions on, vol.56, pp. 4258-4265, 2009.
[4] G. Gupta and M. Younis, "Fault-tolerant clustering of wireless sensor
networks," in Proc. 2003 Wireless Communications and Networking,
2003. WCNC 2003. 2003 IEEE, pp. 1579-1584.
[5] G. Gupta and M. Younis, "Load-balanced clustering of wireless sensor
networks," in Proc. 2003 Communications, 2003. ICC '03. IEEE
International Conference on, pp. 1848-1852.
[6] H. Liu, P. Wan and X. Jia, "Fault-tolerant relay node placement in
wireless sensor networks," in Proc. 2005 11th Annual International
Conference on Computing and Combinatorics, COCOON 2005, August
16, 2005 - August 29, 2005, pp. 230-239.
[7] H. Bin, T. Han and X. Guoliang, "Fault-tolerant relay node placement
in wireless sensor networks: formulation and approximation," in Proc.
2004 High Performance Switching and Routing, 2004. HPSR. 2004
Workshop on, pp. 246-250.
[8] A. Srinivas, G. Zussman and E. Modiano, in MobiHoc06, Florence,
Italy., 2006.
[9] [9] Z. Weiyi, X. Guoliang and S. Misra, "Fault-Tolerant Relay Node
Placement in Wireless Sensor Networks: Problems and Algorithms," in
Proc. 2007 INFOCOM 2007. 26th IEEE International Conference on
Computer Communications. IEEE, pp. 1649-1657.
[10] A. Bari, A. Jaekel and S. Bandyopadhyay, "Optimal Placement of Relay
Nodes in Two-Tiered, Fault Tolerant Sensor Networks," in Proc. 2007
Computers and Communications, 2007. ISCC 2007. 12th IEEE
Symposium on, pp. 159-164.
[11] E. Amaldi, A. Capone, M. Cesana, I. Filippini and F. Malucelli,
"Optimization models and methods for planning wireless mesh
networks," Computer Networks, vol.52, pp. 2159-2171, 2008.
[12] A. So and B. Liang, "Optimal placement and channel assignment of
relay stations in heterogeneous wireless mesh networks by modified
Benders decomposition," Ad Hoc Networks, vol.7, pp. 118-135, 2009.
[13] D. Goldberg, Genetic Algorithms in Search, Optimization, and Machine
Learning, Addison-Wesley Professional, 1989.
[14] M. Gen and R. Cheng, Genetic Algorithms and Engineering
Optimization (Engineering Design and Automation), Wiley-Interscience,
1999.
[15] Q. Xiaofeng and F. Palmieri, "Theoretical analysis of evolutionary
algorithms with an infinite population size in continuous space. Part I:
Basic properties of selection and mutation," Neural Networks, IEEE
Transactions on, vol.5, pp. 102-119, 1994.
[16] G. Rudolph, "Convergence analysis of canonical genetic algorithms,"
Neural Networks, IEEE Transactions on, vol.5, pp. 96-101, 1994.
[17] V. Ravi, B. S. N. Murty and J. Reddy, "Nonequilibrium simulated-
annealing algorithm applied to reliability optimization of complex
systems," Reliability, IEEE Transactions on, vol.46, pp. 233-239, 1997.
[18] W. Lipo, L. Sa, F. Tian and F. Xiuju, "A noisy chaotic neural network
for solving combinatorial optimization problems: stochastic chaotic
simulated annealing," Systems, Man, and Cybernetics, Part B:
Cybernetics, IEEE Transactions on, vol.34, pp. 2119-2125, 2004.
[19] S. Kirkpatrick, C. D. Gelatt and M. P. Vecchi, "Optimization by
Simulated Annealing," Science, Number 4598, 13 May 1983, vol.220,
4598, pp. 671-680, 1983.
[20] M. Yoshikawa, H. Yamauchi and H. Terai, "Hybrid Architecture of
Genetic Algorithm and Simulated Annealing," Engineering Letters,
Letters, 16:3, EL_16_3_11, 2008.
[21] T. Eloranta and E. Mkinen, "{TimGA - A Genetic Algorithm for
Drawing Undirected Graphs},", 1996.
[22] G. De Marco, "MOGAMESH: A multi-objective algorithm for node
placement in wireless mesh networks based on genetic algorithms," in
Proc. 2009 Wireless Communication Systems, 2009. ISWCS 2009. 6th
International Symposium on, pp. 388-392.
[23] ILOG CPLEX 9.1 Documentation. Available at the website
http://www.columbia.edu/%CB%9Cdano/resources/cplex91