Cloud Model and Ant Colony Optimization Based QoS

Routing Algorithm for Wireless Sensor Networks

Di Jian

School of Control & Computer Engineering

North China Electric Power University, Baoding, China
[email protected]

Abstract. This paper presents CMACRO (Cloud Model and Multiple Ant
Colonies Optimization based Routing), a new cross-layer QoS routing algo-
rithm for wireless sensor networks. Basing on the principle of cross-layer
design, the algorithm adapts delay, nodes’ load and link quality as QoS metrics,
and provides differentiated services for real time event-driven data streams and
delay-tolerant periodic sampling data. The QoS routing metrics are regarded as
heuristics correction factors in ant colony algorithm (ACA). The ants are di-
vided into a number of different populations. Through the interaction of phero-
mone between multi populations, the routing algorithm searches for the feasible
paths in parallel and updates the pheromone in time. To overcome the slow
convergence of ant colony algorithm, improvements to control the randomness
of the ants via cloud model are proposed. The simulation results demonstrate
that the routing algorithm can guarantee the real time, reliability and robustness
of wireless sensor networks. It can also achieve the network load balancing and
congestion control mechanism.

Keywords: wireless sensor networks, cloud model, multiple ant colonies algo-
rithm, QoS routing.

1 Introduction
Wireless Sensor Networks (WSN), also known as Ubiquitous Sensor Networks
(USN) ,bring together many disciplines in with the rapidly developing wireless tech-
nology. WSNs [1] are networks based on small size nodes cooperation. Those nodes
are mainly characterized by their low energy consumption, their low cost and, of
course, their wireless communication. The application level is sensitive to the charac-
teristics of network transmission, such as delay, throughput, bandwidth, packet loss
rate and energy cost.
QoS routing technology is an important guarantee of WSN QoS [2]. The role of a
QoS routing strategy is to compute paths that are suitable for different type of traffic
to meet the requirements of various applications. The main problem to be solved by
QoS routing algorithm is multi-objective optimization problem. Ant colony algorithm
(ACA) is a bio-inspired algorithm [3], and it is based on the behavior of an ant colony
in the nature. The distributed nature of network routing is well matched by the multi

Y. Wu (Ed.): International Conference on WTCS 2009, AISC 116, pp. 179–187.

springerlink.com © Springer-Verlag Berlin Heidelberg 2012
180 D. Jian

agent nature of ant colony optimization (ACO). But in the actual operation, there are
two problem to be addressed. First, the algorithm is easy to fall into local extrema.
Second, the algorithm is slow convergence. This is due to the absence of a mechanism
to control the randomness of the decline.
In this paper, through introducing membership cloud models (MCL) to adjust the
random decline degree of ant colony, we firstly propose a membership cloud model
based ant colony algorithm. Subsequently, using this algorithm and adopting the
cross-layer design approach, we present an on demand cross-layer QoS routing algo-
rithm. Our algorithm utilizes routing metrics from different layers to achieve overall
system optimization.

2 Background Information

2.1 Cloud Model

Cloud model, put forward by professor Li Deyi, is a qualitative to quantitative con-

version model [4]. Based on random mathematics and fuzzy mathematics, cloud mod-
el is proposed to descript randomness, fuzziness and the relationship between them in
linguistic value. Cloud model has been successfully applied in intelligent control, data
mining, intrusion detection, fuzzy evaluation and other fields.
Expected value (Ex), Entropy (En) and Hyper-Entropy(He) are digital features
of cloud model, which form the knowledge representation foundation for mapping
between the qualitative and quantitative. A cloud droplet forms from a mapping of
qualitative concepts and quantitative concepts. Uncertainty in the process of its for-
mation, cloud droplet cloud model gives a qualitative ( μ ) idea of the determination
of x. There are positive direction and reverse sub-cloud, which is based on cloud for-
mation mechanism and calculation direction. The cloud generation algorithm can be
achieved by using software, and hardware can also be solidified into implementation,
which is called cloud generator (CG).

2.2 Basic Ant Colony Algorithm (ACA)

The ant colony algorithm is a member of swarm methods family, and it constitutes
some metaheuristic optimizations based on the behavior of ants seeking a path be-
tween their colony and a source of food [5].
Edge selection:
An ant will move from node i to node j with probability
τ ijα (t )ηijβ (1)
pijk ( t ) =
∑τ ijα ( t )ηijβ
j∈N ik

Where
τ ij ( t ) is the amount of pheromone on edge i,j.
ηij is the desirability of edge i , j (a priori knowledge).
α is a parameter to control the influence of τi,j.
β is a parameter to control the influence of ηi,j.
Cloud Model and Ant Colony Optimization Based QoS Routing Algorithm Technology 181

Pheromone update:

τ i，j = ( 1 − ρ )τ i , j + Δτ i , j (2)

Where
τi,j is the amount of pheromone on a given edge i,j.
ρ is the rate of pheromone evaporation
Δτi,j is the amount of pheromone deposited, typically given by
⎧1
⎪ if ant k passededge(i, j)
Δτ = ⎨ LK
k
ij
⎪⎩ 0 otherwise
(3)
where Lk is the cost of the kth ant's tour.

2.3 Adjustment of ACA Parameters Using Membership Cloud Model

The important parameters of the residual factor ρ and the total pheromone information
Q are used to control the randomness of multiple ant colony optimization algorithm.
Membership Cloud Generator (MCG) is used to optimize the residual factor ρ and the
total pheromone information Q so as to update the pheromone adaptively [6]. The
parameters of cloud model are set as the following. x0=0, b=Nmax/3, The initial value
σ σ
of max = 0.
Algorithm 1: Update Q
Step1 initialization, e=0, b=Nmax/3, σmax =σ0, t=0, set Qmax.
Step2 t=t+1;
Step3 x=t- Nmax, Generate U conditions cloud droplets with the aforementioned
digital characteristics;
Step4 Q= Qmax*μ;
Step5σmax =σmax +σdelta;
Step6 return to step 2.
Algorithm 2: Update ρ
Step1 initialization, e=0, b=Nmax/3, σmax =σ0, t=0, setρmax.
Step2 t=t+1;
Step3 x=t, Generate U conditions cloud droplets with the aforementioned digital
characteristics;
Step4ρ=ρmax*μ;
Step5σmax =σmax +σdelta;
Step6 return to step 2.

3 QoS Routing Metrics

3.1 Delay
In wireless sensor networks, the delay between node i and node j comprises the fol-
lowing factors: processing delay denoted as Dpro(i), queue delay denoted as Dqueue(i)
and transmission delay denoted as Dtran(i j).
182 D. Jian

D（i , j ) = D pro ( i ) + Dqueue ( i ) + Dtran ( i , j ) (4)

The processing delay Dpro(i) can be ignored due to the fast speed processing. The
queue Dqueue(i) delay depends on the network load and the queue length. When using
CSMA/CD as MAC protocol, Dtran(i j) can be derived from packet retransmission
times Ri and each retransmission delay dk.
Ri
Dtran ( i , j ) = ∑ d k (5)
k =1

3.2 Node Load

The load value Qi of the node i denotes the queue length for sending datum. Qi can
be obtained from the information table of adjacent nodes. The node i periodically
samples the queue length from the MAC layer interface. The average sampled value
denotes the network load of node i. The queue length of the node i is denoted in
Eq.(6).
n

∑ q i (k )
Qi = k =1
n (6)
Definition 1: The load estimation function of the node i is denoted in Eq.(7).
Li = Q i + ∑Q
j∈ N
i
j (7)

i
Where Li denote the load estimation function of the node i, Q j denotes the queue
length of the node j which is the adjacent node of the node i. The set N is the adjacent
set of the node i. The sensor nodes in WSN cyclical broadcast the “Hello” packets with
the queue length to its adjacent nodes [7]. The load value is used as a routing metric,
which can effectively avoid network congestion and achieve network load balancing.

3.3 Link Quality

The actual transmission links of WSN can’t completely guarantee the packet recep-
tion rate [8], because the transmission lines perennially are exposed to the atmos-
phere, a number of uncertain factors such as climate change and accidental damage
may lead to node failure. Channel interference and data conflict also may increase the
link error rate. The packet reception rate is used as a routing metric in this paper. The
function of packet reception rate is denote in Eq.(8).
Lr + 1
PRR = (8)
Ls + 2
Where Lr denotes the successfully received packet numbers, Ls denotes the total sent
packet numbers. The transmission link with best link quality will be used for data
transmission, so the reliable requirement of the transmission line monitoring system
can be successfully met.
Cloud Model and Ant Colony Optimization Based QoS Routing Algorithm Technology 183

4 Prepare Design of the MCL and ACO Based QoS Routing

Algorithm for WSN
The steps of the new routing optimization algorithm is shown as below.
Step1: Algorithm initialization. The algorithm divides the artificial ants into k groups.
Each group is assigned a corresponding sub-sink node as the destination node. There
are m ants in each group, and each ant carries the ID information of the species, the
source node and the nodes which are passed by the ant, timestamp, an empty stack,
node load Li , packet reception rate PRR i and so on. The ants will be divided into
two types in the network. They are the forward ants and the backward ants. The taboo
list is initialized at first. The forward ant collects the nodes’ information to update the
local pheromone. The backward ant takes the same path as that of its corresponding
forward ant, but in the opposite direction. The backward ant updates the global phe-
romone.
k
Step2: The attract factor and the exclusion factor of pheromone. Ai denotes the ant
agent i in the population of K. The ants in the same population release the same type
of pheromone. Different populations have different types of pheromone [9].

Definition 2: The attract factor. The attract factor α ij is denoted in Eq. (9).
k

τ ijk
α ijk =
∑ τ khk
h⊂∏ i
(9)

Where τij denotes the pheromone trajectory intensity of k at the edge ( i, j ) .

Definition 2: The exclusion factor. The exclusion factor βij is denoted in Eq. (10).
k

∑ τ h
ij
β k
= h≠k
(10)
∑ τ
ij k
ih
h⊂ ∏ i

Where τij denotes the pheromone trajectory intensity of k at the edge ( i, j ) .

Step3: The probability transfer rules. The transfer probability formula is denoted in
Eq. (11).
k
­ H ij / E ijk
D
s
° [ Wij ( t )] u [ Kij ( t )] E ٛ
j  tabu k
° k (11)
Pijs ( k ) ® H ij / E ijk ٛ
D

¦ [ Wihs ( t )] u [ Kih ( t )] E
° htabu k
° 0 ٛ otherwise ٛٛ
¯
Where the attract factor α ij and the exclusion factor βij are given by the formula
k k

(9) and (10). τ ij (t ) denotes the pheromone trajectory intensity of S at the edge eij .
s

The adjustment factor ε is greater than 0. The heuristic function ηij is denotes in
Eq. (12).
184 D. Jian

PRR
η ij = j
(12)
L j × D ij

Where PRR j denotes the packet reception rate of the sensor node j. L j denotes the
network load of the sensor node j. Dij denotes the single-hop delay between the node
i and the node j.
Step4: The dynamic adjustments of parameters based on cloud model. The value of
and Q is adjusted by Algorithm 1, ρ is adjusted by Algorithm 2.
Step5: The local pheromone updating rule of forward ants. Each group of ant colony
releases different types of pheromones. The pheromone trails are updated by the rule
given by Eq. (13).

τ ijk ( t + 1) = (1 − ρ ) × τ ijk ( t ) + ρ × Δ τ ijk (13)

The pheromone residual factor ρ is selected by the qualitative association rules in

cloud model theory. The ant- density model is used to update the pheromone in the
edge ( i , j ) . The number of the pheromone released by the ant is Q. The model is as
below.
⎧Q if ant k passededge(i, j)
Δτ ijk ( t ,t + 1 ) = ⎨ (14)
⎩0 otherwise

Step6: The global pheromone updating rules of backward ants. When the destination
node d is reached, the forward ant generates another backward ant, transfers to it all of
its memory, and dies. The backward ant takes the same path as that of its correspond-
ing forward ant, but in the opposite direction. The pheromone trails are updated by the
rule given by Eq. (15) and Eq. (16).
Q
τ ijk ( t + n ) = (1 − ρ ) × τ ijk ( t ) + ρ × (15)
Rs
D P × L P × HC P
RS = (16)
PRR P

Step7: Set maximum and minimum of the pheromone.

If the pheromone of the path is too high or too low, the algorithm may fall into the
local optimum situation. The numbers of the residual pheromone will be limited at
τk ( t )
[τmin, τmax], and ij will be modified after the ants have completed a tour. The fol-
lowing formula is used to select the threshold[10].
⎧ τ min τ ijk ( t ) ≤ τ min
⎪
τ ij ( t + n ) = ⎨τ ijk ( t )
k
τ min < τ ijk ( t ) < τ max (17)
⎪τ τ ( t ) ≥ τ max
k
⎩ max ij
Cloud Model and Ant Colony Optimization Based QoS Routing Algorithm Technology 185

Step8: When the backward ant reaches the source nodes, the second batch of ants will
be generated. IF the numbers of iterations reach the maximum value-NCmax, the loops
end.

5 Simulation and Performance Analysis

In order to verify the feasibility and effectiveness of the routing algorithm, we make
the comparition of the transmission delay, congestion rate and network throughput
between the CMACRO algorithm and the AODV algorithm. At last the robustness of
the algorithm is verified.
OMNeT++ simulation tool is used and the parameters are chosed as below: The
maximum coverage distance of the node is 1000 miles. The MAC layer is based on
IEEE 80.2.15.4. Multiple rules generator based on cloud model is used to select p and
Q. The numbers of the drops in cloud model is 500. Ex = 0. 5 En = 0. 8 He = 0. 08.
The numbers of the ant colony is 3. The numbers of ants in each group is 100.
ε = 0.85 , ε = 0.85 , β = 2 , ρ min = 0.3 , τ max = τij ( 0 ) , Δτij ( 0 ) = 0 , τ min = τ max 300 , T=5,
k k

NCmax=500.

5.1 The Average Delay of Data Packets

The average delay of data packets is the average value of the waiting delay and the
transmission delay in a fixed period of time. The simulation results of average delay
are shown in Figure 1. We can see from the simulation results that the CMACRO
algorithm is better than the AODV algorithm in terms of average delay.

Fig. 1. Comparison of average delay

5.2 Comparison of Packet Successful Reception Rate

Packet reception rate is defined as the ratio of the data packets delivered to the desti-
nation. As it is illustrated in Figure 2, the CMACRO algorithm is better than the
AODV algorithm in terms of packet reception rate. With the increasing network load,
the packet reception rate of the optimum path becomes lower. However the link quali-
ty of the optimum path obtained by the CMACRO algorithm is better than the AODV
186 D. Jian

algorithm, and the path is more stable. Because the packet reception rate is added to
the transition probability in the CMACRO algorithm, the node will choose the next-
hop node with maximum packet reception rate to transmit the datum.

Fig. 2. Comparison of packet reception rate

5.3 Network Throughput

The node load is added to the transition probability in the CMACRO algorithm, the
node will choose the next-hop node with minimum load value to transmit the datum
in WSN. As it is illustrated in Figure 3, the improved algorithm can guarantee the
network throughput for WSN. The CMACRO algorithm has more significant advan-
tages than the AODV algorithm. For the algorithm can reduce the numbers of the
bottleneck nodes near the sink node, the requirements of the transmission line moni-
toring system can be successfully met.

Fig. 3. Comparison of network throughput

Acknowledgment. This work is supported by the National Natural Science Founda-

tion of China under Grant No. 60974125.
Cloud Model and Ant Colony Optimization Based QoS Routing Algorithm Technology 187

References
[1] Romer, K., Mattern, F.: The Design Space of Wireless Sensor Networks. IEEE Wireless
Communications 11(6), 54–61 (2004)
[2] Akyildiz, I.F., Melodia, T., Chowdhury, K.R.: Survey on wireless multi-media sensor
networks. Computer Networks, 9212960 (2007)
[3] Dorigo, M., Gambardella, L.M., Birattari, M., Martinoli, A., Poli, R., Stützle, T. (eds.):
ANTS 2006. LNCS, vol. 4150. Springer, Heidelberg (2006)
[4] Li, D.Y., Yi, D.: Artificial Intelligence with Uncertainty. National Defense Industry
Press, Beijing (2005)
[5] Dorigo, M., Maniezzo, V., Colorni, A.: Ant system: optimization by a colony cooperat-
ing agents. IEEE Transactions on Systems, Man and Cybernetics part
B:Cybernetics 26(1), 29–41 (1996)
[6] Zhang, Y.-D., Wu, L.-N., Wei, G.: Improved ant colony algorithm based on membership
cloud models. Computer Engineering and Applications 45(27), 11–14 (2009)
[7] He, P.: Study on routing and topology control techniques in mobile Ad Hoc networks.
Xidian University (2007)
[8] Xia, Z., Yu, H., Yang, B.: Algorithm for probabi-listic link selection in wireless sensor
networks using Bayesian estimation. Huazhong Univ. of Sci. & Tech (Natural Science
Edition) 37(2), 40–44 (2009)
[9] Xia, H.: Research on intelligent computation method with application to network optimi-
zation and prediction. Jiangnan University (2009)
[10] Duan, H., Wang, D., Yu, X.: MAX-M IN meeting ant colony algorithm based on cloud
model theory and niche ideology. Journal of Jilin University(Engineering and Technolo-
gy Edition) 36(5), 803–808 (2006)