Congestion control in wireless sensor networks by hybrid multi-objective optimization algorithm

Accepted Manuscript

Congestion Control in Wireless Sensor Networks by Hybrid Multi-Objective Optimization Algorithm Karishma Singh , Karan Singh , Le Hoang Son , Ahmed Aziz PII: DOI: Reference:

S1389-1286(18)30143-9 10.1016/j.comnet.2018.03.023 COMPNW 6450

To appear in:

Computer Networks

Received date: Revised date: Accepted date:

25 August 2017 19 March 2018 22 March 2018

Please cite this article as: Karishma Singh , Karan Singh , Le Hoang Son , Ahmed Aziz , Congestion Control in Wireless Sensor Networks by Hybrid Multi-Objective Optimization Algorithm, Computer Networks (2018), doi: 10.1016/j.comnet.2018.03.023

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

Congestion Control in Wireless Sensor Networks by Hybrid Multi-Objective Optimization Algorithm Karishma Singh1, Karan Singh1, Le Hoang Son 2*, Ahmed Aziz1 School of Computer & Systems Sciences, Jawaharlal Nehru University, New Delhi, India [email protected], [email protected], [email protected]

1

VNU University of Science, Vietnam National University, Vietnam [email protected]

CR IP T

2

*: Corresponding author. Tel.: (+84) 904.171.284. Address: 334 Nguyen Trai, Thanh Xuan, Hanoi, Vietnam

CE

PT

ED

M

AN US

Abstract: In this paper, a new congestion control algorithm for Wireless Sensor Networks is proposed. The existing algorithms for this problem have high complexity and power usage due to retransmission with congestion control being carried out by finding the optimal rate through a simple Poisson process. Retransmission of the colliding packets causes wastage of energy since wireless sensor network has limited battery. It has been realized that heuristic based methods offer better rate than the simple Poisson process. Besides, energy of the nodes was not considered in the fitness function of the related algorithms, which can lead to node failure when low energy nodes are used for sending high amount of packets. In order to handle those limitations, we propose a congestion control algorithm based on the multi-objective optimization algorithm named PSOGSA for rate optimization and regulating arrival rate of data from every child node to the parent node. A multi-objective optimization function taking into consideration the energy of the node in its fitness function is used. The priority based transmission is enabled as the optimization approach regulates the arrival rate on the basis of priority: output available bandwidth and energy of the child node. To mitigate the congestion, adjustment of rate to optimum value is used. The new algorithm is implemented in MATLAB R2016a and compared against the existing Cuckoo Search (CS) and Adaptive Cuckoo Search (ACS) algorithms. Simulation results prove that proposed mechanism has better results than the existing approaches.

AC

Keywords: Wireless sensor networks; Congestion control; Congestion levels; PSOGSA; Optimization.

ACCEPTED MANUSCRIPT 2

1. Introduction

AC

CE

PT

ED

M

AN US

CR IP T

In real applications where several nodes in a Wireless Sensor Network (WSN) send data to a single sink node at the same time period, there are chances of congestion in the network [1]. When a sensor node accepts data packets at higher rate than its capability to transmit, extra data needs to be stored in buffer [22]. Due to limited availability of space, buffer becomes full and data packets (new or old) have to be dropped as a result of congestion [23]. As such, important data packets may be dropped, which can essentially nullify the purpose of sensor networks [27-28]. In case of multi-hop wireless Ad-hoc networks, a single routing metric is not always the best available solution. This is because using the same single route for multiple communication sessions may result in performance degradation in a network in the form of severe information loss due to congestion. Due to the usage of multiple paths, the problems such as route coupling, collision, and the channel access rate might occur, which may decrease the performance of the network. Congestion in Wireless Sensor Networks has following negative effect on performance such as decrease in throughput and delivery ratio, and increased delay and per-packet energy consumption [2]. Hence, it becomes necessary to give attention to the problem of congestion in the sensor network to provide the needed delivery ratio for WSN applications, and to prolong the network lifetime [25-26]. In order to handle those limitations, we propose a new congestion control algorithm based on the multi-objective optimization algorithm named PSOGSA [3] for rate optimization and regulating arrival rate of data from every child node to the parent node. Hybrid algorithms have the capability to avoid local optimums with faster convergence than other heuristic approaches. Meta-heuristic algorithm uses pattern matrices to give random solutions. In Particle Swarm Optimization (PSO), swarm is referred by a pattern matrix and each pattern corresponds to artificial particle. A pattern is considered as an artificial nest in the Cuckoo Search (CS) algorithm. Gravitational Search Algorithm (GSA)- a powerful optimization algorithm in many applications uses exploration. The proposed algorithm is a combination of PSO [4] and GSA [5] to integrate the strength of both algorithms. It uses two solution strategies namely exploration and exploitation. The exploration process succeeds in enabling the algorithm to reach the best local solutions within the search space. The exploitation process expresses the ability to reach the global optimum solution which is likely to exist around the local solutions obtained. The hybrid PSOGSA combines the ability of global search in PSO with the local search capability of GSA. Compared with the other heuristic approaches, the proposed hybrid PSOGSA gives faster convergence and better solution quality irrespective of number of iterations (please refer to Refs. [31-35] for details). Specifically, PSOGSA is an efficient algorithm as the fitness or the quality of solutions is considered in each iteration. Agents having more attraction force, and close to the good solutions attract the other agents. When all agents are close to a good solution, they move slowly. PSOGSA uses a variable as memory to store the best solution found among all agents. Thus each agent can keep track of the best solution so far and move toward it. This concept has been used to reduce the congestion problem as the optimization algorithm avoids the congestion in the congested network by decreasing the average arrival rate of data from child nodes to the parent node. Using PSOGSA, the optimization algorithm yields the new arrival rate from the child node which is less than that of the arrival rate of the last period. The optimization algorithm gives the fitness function based on the network congestion parameters based on which the control is provided. The selection of the new arrival rate for the child node is set using the selected fitness function value. To sum up, the new contributions in comparison with the previous works are shown below: (a) The new algorithm handles the limitation of previous work efficiently: In the previous works, the rate optimization algorithm has disadvantage of the complexity factor and the power usage due to the retransmission. Retransmission of the colliding packets causes wastage of energy since wireless sensor network has limited battery. The congestion control was carried out by finding the optimal rate which is found out after solving the optimization problem utilizing a simple Poisson process. In PSOGSA, this is done by a hybrid multi-optimization algorithm that allows the arrival rate of data from the child nodes to be within the transmission or service rate of the parent node; thereby helping in the congestion control. Whenever the parent node traffic exceeds the queue size,

ACCEPTED MANUSCRIPT 3

AN US

CR IP T

the PSOGSA took over its part neutralizing the congestion. Thus, the new algorithm would achieve better rate suggestion than the simple Poisson process. (b) Optimal Fitness Function: In the existing congestion control approaches, the energy of nodes is not considered in its fitness function, which can lead to node failure when low energy nodes are used for sending high amount of packets. The proposed algorithm solves this limitation by regulating the arrival rate of data from every child node to the parent node by using a multi-objective optimization function by taking into consideration the energy of the node in its fitness function. A new fitness function to evaluate the share rate of the child node to be within the service rate of the parental node is presented. The maximal fitness function chooses the new share rate which is less than that of the maximal share rate in previous iteration of the data transfer. The priority based transmission is enabled as the optimization approach adapts the arrival rate from every child node on the basis of its priority, output available bandwidth and energy of the child node. The priority for the child and parent node in the sensor node cluster is defined. Depending on the priority of the node, the data transmission occurs in the WSN. The new algorithm will be experimentally validated and compared against the existing Cuckoo Search (CS), Adaptive Cuckoo Search (ACS), and the PSOGSA algorithms. The rest of the paper is organized as follows. Section 2 presents the related works. Section 3 presents the theory of Particle Swarm optimization (PSO) and Gravitational Search Algorithm (GSA). Section 4 describes the problem statement. Section 5 presents the model and approach. Section 6 presents the evaluation and results. Lastly, Section 7 highlights the conclusions and further works.

2. Related Works

AC

CE

PT

ED

M

In order to handle the congestion problem, various algorithms have been proposed. Wan et al. [9] proposed the Congestion Detection and Avoidance (CODA) algorithm for congestion control and avoidance by applying three mechanisms: detection of congestion, notification of congestion by applying hop-by hop backpressure notification, and congestion control using closed-loop, multi-source AIMD-like traffic control to mitigate congestion. Congestion detection is used to notice the occurrence of congestion in any area in the network so that rest of the mechanisms can be activated. Buffer occupancy and channel status or channel loading conditions are used to detect the congestion. CODA uses AIMD (Additive Increase multiplicative Decrease) [10] for congestion control. Ee and Bajcsy [11] proposed congestion control and fairness (CCF) for many-to-one routing in which a scalable and distributed algorithm is used to eliminate congestion and ensures transparent packet delivery to a base station or central node. In addition, when it has received the equal number of the packets then fairness is achieved. Trickle [12] is a congestion control protocol which is used to propagate and maintain code updates. It broadcasts an update only when a mote hears an older summary than its own. Each mote acquires less tickle of packet only when algorithm controls the send rate. It has greater applicability and used to disseminate any sort of data. Wang et al. [13] proposed a congestion control protocol in WSN Priority based Congestion Control (PCCP). Node priority based on their function or location is used to reflect the significance of each node present in network. This protocol uses the inter arrival time of packet along with service time of packet for measuring a parameter that defines to congestion degree and also imposes hop-by-hop control which is based on degree of measure congestion and priority index of the node. Ahmad et al. [14] proposed Congestion Avoidance and Fairness (CAF) that uses available queue size or buffer occupancy of downstream nodes and characteristic ratio of number of downstream nodes and upstream nodes. CAF provides fairness and efficient load balancing by monitoring queue sizes or buffer occupancy of downstream nodes. Hybrid Congestion Control Protocol (HCCP) [15] is another algorithm that detects congestion using incoming data rate, outgoing data rate and buffer size of nodes. In HCCP, if congestion degree is greater than zero, congestion does not occur in the next time interval and vice versa. Multi-event Congestion Control Protocol (MCCP) [16] supports event reporting of three types such as prioritized multiple event reporting, per-node fair event reporting and general event

ACCEPTED MANUSCRIPT 4

AC

CE

PT

ED

M

AN US

CR IP T

reporting. Depending on the selected event reporting mode, MCCP helps to mitigates congestion and provides outcome with respect to selected event reporting mode. Sergiou et al. [17] proposed Dynamic Alternative Path Selection Algorithm to measure the congestion by increasing capacity but condition is to maintain performance requirements. Adaptive cuckoo search based optimal rate adjustment (ACSRO) [7, 18] was used for the congestion avoidance and control. The share rate of anode is regulated by rate adjustment to minimize the congestion which enables the sensor network to recover from the packet loss and detect incipient congestion. Antoniou et al. [19] proposed Flock-CC based on the bird flocking behavior. Rezaee et al. [6] proposed Optimized Congestion Management Protocol with high rate optimization when considering packets loss and managing queues. REFIACC [24] prevents the interferences and ensures a high fairness of bandwidth utilization among sensor nodes by scheduling the communications. The congestion and the interference in inter and intra paths hot spots are mitigated through tackling into account the dissimilarity between links’ capacities at the scheduling process. Linear programming is used to reach optimum utilization efficiency of the maximum available bandwidth. REFIACC has been evaluated by simulation and compared with two pertinent works. The results show that the proposed solution outperforms the others in terms of throughput and reception ratio (more than 80%) and can scale for large networks. In [25], a method based on particle swarm optimization and fuzzy clustering is proposed to handle the problem of network disconnect. Experimental validation shows that the proposed method is better than the existing ones. Besides, some other congestion control algorithms were applied to mitigate the congestion such as BDCC [20] and DAIPaS [17]. The relevant algorithms find difficulty of extra overhead to the already heavily loaded environment that leads to resource depletion. Due to the undesirable situation of congestion in network, various congestion control algorithms were applied in order to mitigate congestion, such as BDCC [20] and DAIPaS [17]. However, these algorithms give additional overhead to the already heavily loaded environment. In BDCC, Even though the dynamic selection of the router improves the performance, the priority selection completely depends on the network leading to the improper selection. The quality of service was also inefficient. In DAIPaS, the minimal traffic overhead based congestion control scheme was proposed. In this regard, the complexity in the computation makes it unfit for the transport protocol. In Optimized Congestion Management Protocol [6], disadvantages are complexity and power usage due to the retransmission. Retransmission of the colliding packets causes wastage of energy since wireless sensor network has limited battery. Congestion control was carried out by finding the optimal rate found by a simple Poisson process which has lower rate suggestion than a heuristic method. Besides, energy of the nodes was not considered in the fitness function of the algorithm ACSRO, which can lead to node failure when low energy nodes are used for sending high amount of packets. In [7], simulations over the existing rate optimization schemes (Systematic Share Rate reduction (SS), Cuckoo Search (CS) [18], and Optimal Rate Adjustment scheme (ORA) [32]) were performed to analyze performance. The results showed that the proposed ACSRO scheme is more efficient than the existing rate optimization schemes for congestion management. They suggested the rate adjustment can be performed with the hybrid search algorithm by combining multiple optimization algorithms which inspired the proposed work. In [31], hybrid PSOGSA and Cuckoo Search (CS) algorithm have been implemented for global optimization of Reactive Distillation (RD). In this regard, the hybrid PSOGSA gives faster convergence and best solution quality irrespective of number of iterations. 3. Preliminary 3.1. Particle Swarm Optimization Particle Swarm Optimization (PSO) [4] is guided by fish schooling and bird swarm intelligence. In context of fish schooling [8], all individuals evolved with competition and cooperation among themselves. Every agent adapts flying by learning from own experience and experience of its companions. A number of particles are considered flying in the search space synchronously to

ACCEPTED MANUSCRIPT 5

discover the best solution. Each of them is considered as a potential solution to a problem. They all have focus on the best solution coming in their paths. Agent takes into consideration its own best solutions and also the best solution found so far. We consider the system with N particles (i=1, 2, … , N)in D-dimensional space. Variables used for model of PSO are listed in Table 1. Initially, particles are placed randomly. Velocities of all the particles are computed by equation (1). After calculating velocities, the particle flies to the new position that is calculated using equation (2). Performance of every particle is calculated using a predefined fitness function. (

)

( )

( ))

( )

( )

(

( ))

)

(1) (2)

CR IP T

(

(

where

- An inertia weight function that balances the local search and global search, and - positive constants which are cognitive and social parameters respectively, and - random number between 0 and 1.

AN US

Table 1. List of Variables used in PSO

Variable Name

Description

Pos

Position of each particle

Vel

Velocity of each particle

Pbest Gbest

Previous best position of particle Best solution in whole population

ED

M

In equation (1), ( ) provides ‚exploration‛capability. The second part ( ( )) is the ‚cognition‛ part which represents local search capability or private ( thinking and third part ( ))is the ‚social‛ part which shows the global search capability or collaboration of all the particles. 3.2. Gravitational Search Algorithm

AC

CE

PT

Gravitational Search Algorithm (GSA) [5] is motivated by the basic physical phenomena of Newton’s theory. It considers a number of agents or candidate solutions placed in the search space which have the mass proportional to the fitness function value. The heavy masses are having more attraction force, and they attract other masses proportional to their distances. The mathematical model of GSA is discussed here. Consider the system of N agents or masses (i=1, 2, … , N) in D-dimensional space. Variables used for model of GSA are listed in Table 2. Force acting from mass ‘j’ on mass ‘i’, at time ‘t’, is defined in equation (3): ( ) ( ) (3) ( ) ( ) ( ) ( )) ( ( ) The total force that acts on agent ( )

is calculated as in equation (4): ∑

( )

The G is calculated using initial value (G0) and time (t) using equation (5): ( ) ( )

(4)

(5)

The acceleration of ith agent at specific time t in Dth direction is computed as in equation (6): ( ) (6) ( ) ( ) The velocity of the agents is computed as given in equation (7):

ACCEPTED MANUSCRIPT 6

(

)

( )

( )

(7)

The position of the agents is computed as in equation (8): ( ) ( ) ( )

(8)

Table 2. List of Variables used in GSA

Description

Pos

Position of each particle

Vel

Velocity of each particle

Pbest

Previous best position of particle

Gbest

Best solution in whole population

G(t)

Gravitational constant at time t

Maj

Active gravitational mass of agent j

Mpi

Passive gravitational mass of agent i

Mii

Inertial mass of agent i

F

Force

Acc

Acceleration

AN US

CR IP T

Variable Name

Constant

Euclidian distance between agents i and j Random number between 0 and 1

AC

CE

PT

ED

M

R

Figure 1. Steps of PSOGSA [3]

ACCEPTED MANUSCRIPT 7

3.3. PSOGSA PSOGSA [3] is a combined form of the PSO and GSA algorithm to aggregate the strength of these algorithms. Here the ‚social thinking ( ) capability‛ in PSO is combined with the ‚local search capability‛ of GSA. Thus, velocity is computed as in equation (9): (

)

( )

( )

(

( ))

The positions of particles are computed by equation (10): ( ) ( ) ( )

(9) (10)

CR IP T

Variables used for model of PSOGSA are listed in Table 3. Figure 1 represents the steps of PSOGSA. Firstly all agents are initialized randomly as candidate solutions. After getting done the initialization, we calculate the Gravitational force, resultant forces among agents and gravitational constant using equations (3), (4), and (5) respectively. Then acceleration is computed as (6). After the accelerations are calculated and the best solution so far is updated, the velocities and positions of all agents can be computed using equations (9) and (10). The task of updating the velocities and the positions will stop by meeting an end criterion.

Variable Name

Description

Pos

Position of each particle

Vel

Velocity of each particle

Gbest

Best solution in whole population

W

Weight function

K

M

AN US

Table 3. List of Variables used in PSOGSA

Positive constant

R 4. Problem Definition

Acceleration Random number between 0 and 1

ED

Acc

AC

CE

PT

In a WSN, the reliable data transfer happens from nodes to the base station. The child nodes are the nodes from which the data packets are jointly collected in the parent node from where the data is transferred to base station aiding in the application of the WSN. An example of the cluster with a base station, three parent nodes and seven child nodes are shown in Figure 2. When the arrival rate from the child nodes exceeds the transmission rate from the parent nodes, the data packets send by the child nodes are made to wait in the queue based on the physical queue size of the parent node. When the data packets exceed the size tolerable by the queue of the parent node, congestion occurs. The main factor for the congestion is the increase in the arrival rate over the packet transmission rate of node. Consider is the child node, is the parent node and is the data packets send over the node. The nodes considered here for the representation are the sensor nodes as in Figure 3. Let be the arrival rate of the data from child node to parent node and be the transmission rate from the parent node to base station respectively. Figure 3 represents the congestion behavior due to the dissimilarity between the arrival rate and the transmission rate. Here, the conditions are as follows: > :Congestion occurs < :Congestion free transmission Let us consider an example, and with the sending rate of one data packet

be the three child nodes sending the data packets at a time. The service or transmission rate of the

ACCEPTED MANUSCRIPT 8

AN US

CR IP T

parent node is two packets at a time. The parent node will get one data packet from each child but it can send two data packet to the base station. At the time of the transmission, at the first instance, three data packets are arrived at parent node from the three child node. The parent node receive it, but the queue length size is only 2. Thus, one of the data packets came from the child node is queued leading to the congestion.

AC

CE

PT

ED

M

Figure 2. Clustering in WSN

Figure 3. Congestion in WSN

ACCEPTED MANUSCRIPT 9

- Parent Node - Child Node - Data Packet to send - Arrival rate of data from Child Node to Parent Node - Transmission Rate from Parent Node to Base Station

CR IP T

At the second instance, three packets are sent; two of them are queued due to the non-availability of the queue size leading to the congestion. This is because the arrival rate is greater than the transmission rate of the node. The rate adjustment optimization reduces the arrival rate of the child node reducing the congestion. Figure 3 represents the congestion behavior in the WSN due to the arrival rate which is far greater than the packet transmission rate of the parent node. 5. Model and Approach

The detailed discussion about the proposed congestion control mechanism for WSNs based on the new hybrid multi-objective optimization (PSOGSA) which combines Particle Swarm optimization (PSO) and Gravitational Search Algorithm (GSA) is discussed in this section.

AN US

5.1 Architecture

AC

CE

PT

ED

M

The architecture of the congestion control using the PSOGSA is given in Figure 4. As arrival rate increases considerably greater than the transmission rate, congestion occurs. If the congestion tends to happen then the level of the congestion is found out using the total size of the queue. The congestion level corresponds to the stage at which the incoming packets start to congest. The congestion level is set within a threshold value. As congestion level increases beyond threshold value, the congestion tends to happen so the rate optimization procedure is called.

Figure 4. Overall architecture of proposed PSOGSA scheme

ACCEPTED MANUSCRIPT 10

AN US

CR IP T

By providing the rate optimization to the congested level, the congestion is controlled in the sensor networks. The rate optimization is nothing but decreasing the arrival rate of data from the child nodes. Optimization algorithm yields the new arrival rate for the child nodes which are less than that of the arrival rate of the last period. The congestion control depends on the parameters like the data arrival rate from the child node, available bandwidth of the node and transmission rate of the parent node. These parameters are optimized in a way such that the arrival rate of data from the child node is reduced. The optimization is done using the PSOGSA based algorithm. It is combined form of the PSO and GSA algorithm to aggregate the strength of both algorithms. We hybridize PSO with GSA by combining the functionality of both algorithms. The fitness function provided by the optimization algorithm gives away the portion of the child node for sending the node data. The new arrival rate adjusted by the proposed PSOGSA optimization algorithm controls the congestion which is rate-adjusted i.e. arrival rate of data is reduced than that of the previous arrival rate of the nodes and the data transport takes place with the adjusted arrival rate. The congestion notification function is fed into the packets’ header part. There by, the extra control messages that are send for the congestion control are avoided which increases the energy efficiency, and thus congestion in the nodes is controlled. The rate optimization procedure goes on continuously aiding in the congestion free transport of the data packets in the WSN. 5.2 Congestion Control

In the proposed approach, congestion control is done as follows:

If∑

or

M

a) Congestion Status: The congestion status is based on the threshold value chosen. The condition to check for the status of the congestion is given in equation (11). where,

(11)

AC

CE

PT

ED

is the size of nth virtual queue that relates to nth child node. is the priority of the child node. is the physical queue size. shows the share of virtual queue in physical queue PQ for child l. If virtual queue total length becomes more than or if for every neighbor of R (‚R is half of neighborhood number m/2‛), the virtual queue size exceeds 95 percent of its portion, the congestion occurs. b) Congestion Level: Once the presence of the congestion in the network is confirmed by the status of the congestion, the congestion level is to be found out. The child node congestion level and overall node congestion level is calculated. The condition for the child node congestion level detection is computed as given in equation (12). The overall node congestion level is calculated as given in equation (13). then

else

(12)

where is the physical queue (PQ) portion of l st child node in parent node k in [0,1]. is used for the rate adjustment. Node congestion level of k sensor nodes may be computed as given in: (13) ∑ shows the available bandwidth (ABW) in the node (ABW = 1−nClevel). If the value of the congestion level increases, the available bandwidth decreases. When the value of exceeds a threshold, the corresponding node experiences the congestion. This implies if the data arrival rate is more than data transmission rate ( ), the congestion occurs at the node. The overall current congestion at the node k is denoted using the value.

ACCEPTED MANUSCRIPT 11

Congestion Notification: The additional overhead of transmission of control messages can be reduced by using the Implicit Congestion Notification as the information related to congestion piggybacks in header part of the packets. d) PSOGSA based Rate Optimization for congestion control: The congestion occurs at the parent node l in Figure 3 because the transmission rate of parent node is less than that of the total average rate of the child nodes. The optimization problem is induced to reduce the congestion problem. The optimization algorithm avoids the congestion in the congested network by decreasing the average arrival rate of data from child nodes to the parent node. By doing so, the data service rate of the parent node is made to exceed the total average rate of the child node. The optimization algorithm yields the new arrival rate from the child nodes which are less than that of the arrival rate of the last period. The optimization algorithm gives the fitness function based on the network congestion parameters based on which the control is provided. The selection of the new arrival rate for the child node is set using the selected fitness function value. e) Congestion Control Solution: The solution of the proposed congestion control scheme is the new arrival rate. In the WSNs, the multiple sensor nodes for the data transfer are available. The simultaneous reception from all child nodes gets congested at parent node. At the time, the physical size of the queue is overflowed, limiting the further acceptance of packets, leading to the loss of packets. During retransmission, the lost packets are traced back but the power usage of the network exceeds, resulting in poor efficiency of the network.

AN US

CR IP T

c)

AC

CE

PT

ED

M

Let us consider the following notations: be the parent node, be the child node, be the arrival rate before adjustment optimization algorithm and be the new adjusted arrival rate. At the congestion state, average of arrival data rate from child node is greater than the transmission rate of parent node. Once the status of the congestion is checked, the congestion control is done by the rate optimization using PSOGSA. The arrival rate of the child node is reduced within the threshold limit by the optimization algorithm. The new arrival rate is the rate adjusted arrival rate. By selecting the new arrival rate for sharing the data packets, the congestion is controlled by the proposed mechanism. Figure 5 shows the solution encoding of the proposed rate optimization system as example of one parent and two child nodes.

Figure 5. Congestion Control solution

ACCEPTED MANUSCRIPT 12

Here, and are the child nodes with the arrival rate . is the parent node accepting the data packets from the child node and sends it to the base station based on its transmission rate . At the arrival rate , the network is congested because the net arrival rate exceeds the transmission rate of parent node. Thus, the rate optimized arrival rate is provided based on the optimization algorithm and it provides a congestion free process in the sensor network. Fitness function to choose best solution: Fitness function is used for the rate optimization is described in this section. In the proposed fitness function, multiple Quality of Service (QoS) parameters are considered for the maximization of the fitness function. The QoS parameters indulged in the proposed fitness function are the data arrival rate, available bandwidth, congestion, transmission rate and energy. By keeping these six constraints, the objective of the maximum optimal reachability was reached in the proposed fitness function. The proposed fitness function considers these six parameters which are identified as most important for controlling the congestion of the network. Also, adjusting the arrival rate may affect QoS. If we consider the maximum number of parameters within the fitness function, QoS can be retained with optimal arrival rate. This is the reason we include six parameters for finding the fitness function. The fitness function must be maximal for the new arrival rate evaluation. The fitness function considering different QoS parameters is given by the equation (14):

AN US

CR IP T

f)

(

)

(14)

PT

ED

M

The first parameter in the fitness function is the arrival rate measured in number of packets per second. The objective function considered with the arrival rate is to set the new arrival rate of the data from child node less than previous arrival rate of the child nodes (at the time of the congestion). The value of the arrival rate is smaller than the value of the transmission rate, thereby avoiding the congestion in the network. When the difference in the rate exceeds, the value of the new arrival rate is minimal. By subtracting the value obtained from the overall difference between the new arrival rate and the old arrival rate with the old arrival rate in the denominator by one, the new arrival rate value that is less than the previous arrival rate value is attained. The first maximal factor in the fitness function regarding arrival rate of the child node is given as calculated using the equation (15), (

(

)

)

(15)

AC

CE

where is old arrival rate and is new arrival rate. The second parameter considered for the fitness evaluation is the available bandwidth ABW measured in byte per second. The bandwidth of the node at which the arrival rate reduction is held to control the congestion, depends on the value of the new arrival rate to be reduced for the optimal fitness reachability. The bandwidth factor desires to be the difference between the available bandwidth of the node and the new arrival rate value. The second factor in the fitness function is based on the available bandwidth of the child node, and it is calculated using the equation (16), (

)

(16)

where ABW is the available bandwidth of the node and is the new arrival rate of the node. The third parameter considered in the fitness evaluation function is the transmission rate measured in number of packets per second. Transmission rate of the parent node should be maximal for the increased performance of the data transmission in the wireless sensor network. The objective was to maximize the transmission rate function. It is achieved by subtracting the value of the difference in the new arrival rate and old arrival rate with the transmission rate of parent node in the

ACCEPTED MANUSCRIPT 13

denominator by one. The range of the transmission rate for the maximal function is attained. The equation for the transmission rate evaluation in the objective formulation is given below: (

(

)

(17)

)

)

(18)

AN US

(

CR IP T

where is the old arrival rate, is the new arrival rate and is the transmission rate of the parent node. The fourth parameter considered for the fitness evaluation is the congestion measured in number of packets per second. The congestion factor should be minimal in the maximal fitness function. The congestion must be reduced for the efficient performance of the sensor networks. The lower value of the congestion provides better objective formulation function. The congestion depends on the status of the congestion level remunerating the congestion control. By the ratio between the factor with the difference from the overall congestion level with one and the arrival rate, the congestion function is made minimal in the proposed fitness function. The value of the overall congestion level is minimized by subtracting the value over the range value 1, which is the main factor in the congestion minimization. The congestion function is calculated using the equation (18),

M

where is the overall congestion level. The fifth parameter considered for the fitness evaluation is the queue length measured in number of packets lost per second. The queue length in the transmission network should be minimal for the improved performance. It depends on the difference between the two factors considered with the physical queue length of the parent node. The function regarding the queue length is calculated using the equation (19), (

)

(

)

(19)

)

ED

(

AC

CE

PT

where is the physical queue size, is the virtual queue length, is the old arrival rate and is the new arrival rate. The sixth and final parameter in fitness function was added is sum of energies of links in the path. The energy in the transmission path network must be maximal for the improved performance. It depends on the energy needed to send and receive data between two nodes through a path and number of packets sent. Suppose the total number of nodes is there. If we consider k clusters, then on average there will be n/k total nodes per cluster, where each cluster having one parent node and (n/k) - 1 Child nodes. The energy consumption for a single child node EChild is only for transmission of m bits to Parent node. Thus, the energy used in each child node [22] is given in equations (20) and (21): (

)

( )

(

)

(20) (21)

Each parent dissipates energy EParent, in reception of signals from all the child nodes of that cluster, aggregation of the signals, and transmission of the aggregate signal to Base Station. Hence, the energy dissipated in the parent node [21] is given in equations (22-24): ( ( (

) )

(

) ( )

(22)

) (

)

(23) (24)

ACCEPTED MANUSCRIPT 14

where EDA is energy used by Parent for data aggregation. Now the energy consumption in a cluster given by [21]is calculated by equation (25): (

(

)

)

(25)

Based on this maximal fitness function, the new arrival rate for the congestion control mechanism is chosen. The new arrival rate provided by the fitness formulation function is selected using the optimization algorithm.

CR IP T

5.3 PSOGSA based Optimization Algorithm

AC

CE

PT

ED

M

AN US

The proposed algorithm used for the optimization of the rate adjustment is discussed in this section. Input: Initial population of n agents ( ). Output: the best possible solution in consecutive iteration is ranked based on the fitness function. The general steps for algorithms are as follows: Step 1: Initialization All agents are initialized randomly. Every agent is considered as a candidate solution. Let us assume that the initial population is the new arrival rate to be chosen . Let the initial population of n agents be ( ). Step 2: Find best solution by PSOGSA using iterative procedure The number of the iterations is defined by the user for the optimal best solution selection. In every iteration, best solution will be updated. From all the agents, a solution i.e. arrival rate is chosen and the fitness is evaluated for the selected one. If the fitness function of the chosen arrival rate is less than that of the PSOGSA selected arrival rate (e.g. ), the arrival rate of the PSOGSA replaces the chosen solution from the agents. From all the agents A, a random agent is chosen say . Calculate the fitness function of . If ( ) ( ) then replace by the solution generated by the PSOGSA. Step 3: Worst agent rejection The solution with minimal fitness value is neglected. In the proposed case, the arrival rate which is greater than that of the previous arrival rate is omitted using the worst case scenario. In this way, worst agents are abandoned. Step 4: Ranking The best possible solution in consecutive iteration is ranked based on the fitness function. Keep the best solution and rank it. Step 5: Iteration Iterate until tolerance is reached. Based on the rank value, the best possible solution is chosen. Thus, new arrival rate providing a better congestion control mechanism in the WSNs. By the continuous iteration of PSOGSA optimization algorithm, the best optimal arrival rate i.e. new arrival rate for the congestion free data transfer in WSN is obtained. The new arrival rate provided by the PSOGSA algorithm is set as the arrival rate of the data from the child node to the parent node. The new arrival rate obtained is found to be less than the arrival rate at the time of the congestion in the network. Thereby, congestion in the WSNs is avoided by the proposed PSOGSA based scheme. PSOGSA is an efficient algorithm as the fitness or the quality of solutions is considered in each iteration. Agents having more attraction force, and close to the good solutions attract the other agents. When all agents are close to a good solution, they move slowly. PSOGSA uses the variable ( ) as memory to store the best solution found among all agents. Thus each agent can keep track of the best solution so far and move toward it. Figure 6 represents the flowchart of the proposed PSOGSA scheme for the congestion control in the WSN.

ACCEPTED MANUSCRIPT 15

AC

CE

PT

ED

M

AN US

CR IP T

The initialization is the simulation of the WSN. After the initiation, because of large number of the nodes in the network, priority based transmission is enabled. The optimization approach regulates the arrival rate on the basis of priority: output available bandwidth and energy of the child node. The priority for the child and parent node in the sensor node cluster is defined. Depending on the priority of the node, the data transmission occurs in the WSN. During the transmission, as the packets are sent over the node, the status of the congestion in the network based on the queue size of the parent node is found out. When the arrival rate from the child nodes exceeds the transmission rate from the parent nodes, the data packets send by the child nodes are made to wait in the queue based on the physical queue size of the parent node. When the data packets exceed the size tolerable by the queue of the parent node, congestion occurs. Non-availability of the queue size leads to the congestion. In status estimation, if the possibility for the congestion is not briefed, then the transmission is congestion free. But if the status of the congestion is true, then the rate optimization is performed using the PSOGSA to adjust the arrival rate of the data from the child node. The new arrival rate is chosen by the proposed new fitness function and by the application of the proposed PSOGSA algorithm, the optimal arrival rate is reached. Thus, the congestion in the WSN network is avoided and further controlled by the optimized rate adjustment.

Figure 6. Flowchart of proposed PSOGSA

ACCEPTED MANUSCRIPT 16

6. Evaluation This section discusses the simulation setup and simulation results. The parameters considered for comparison and the actual simulation results are given here. 6.1 Simulation Setup

AC

CE

PT

ED

M

AN US

CR IP T

This section describes the setup for the simulation which has been carried out by a library for wireless transmission in the MATLAB. It uses Low-energy adaptive clustering hierarchy ("LEACH")1- a hierarchical protocol in which most nodes transmit to cluster heads, and the cluster heads aggregate and compress the data and forward it to the base station (sink). Each node uses a stochastic algorithm at each round to determine whether it will become a cluster head in this round. The goal of LEACH is to lower the energy consumption required to create and maintain clusters in order to improve the life time of a wireless sensor network. Traffic Generated as Poisson Process; this expresses the probability of a given number of packets occurring in a fixed interval of time with a known average rate and independent of the time. Poisson distribution is used to simulate congestion in the proposed work.

Figure 7. Existence of various paths to go from one node to another

1

https://www.mathworks.com/matlabcentral/fileexchange/48162-leach--low-energy-adaptive-clustering-hierarchy-protocol-

ACCEPTED MANUSCRIPT 17

A traffic generation model is a stochastic model of the data sources or traffic flows in a communication network. We have taken the sensor network with 1000m X 1000m square sensor field and the network was simulated with 100 sensor nodes distributed randomly in sensor field shown in Figure 7. Simulations were executed for 10 simulator seconds. Table 4 lists the parameters for the simulation. Table 4. Simulation parameters [7, 18]

Value

Description

Simulation Area Time

1000x1000 m 10 s

Simulation Area (mxm) Simulation Time (sec)

N

100

Number of nodes

R

200 m

Transmission range of each node (m)

Sigma

100Mb/s

mean traffic rate (Mb/s)

Theta

0-45

Standard deviation in traffic

Low_traffic

1.00E+06MB/s

Lower range of traffic demand through nodes in MB/s

High_traffic

1.00E+07MB/s

Upper range of traffic demand through nodes in MB/s

AN US

CR IP T

Variable Name

6.2 Simulation Results

AC

CE

PT

ED

M

We have simulated PSOGSA protocol using MATLAB simulator over windows operating system. To do performance evaluation of CS, ACS and proposed PSOGSA, parameters used are discussed in this section: 1) Throughput: The total network throughput is defined as the data packet receiving rate of the sink (bits/s) over the total bandwidth of the sink (bits/s). 2) Delay: The amount of the time elapsed to send the data packet over the wireless sensor network from the source node to the sink node is called delay. It is measured in seconds (s). 3) Normalized packet loss: The packet loss is the difference in the number of the data packets send and received over the time interval in the network. 4) Sending rate: Sending rate is the rate at which the data packets from the source node/child node are sent to the parental node, measured in bits/s. For every time instance, the share rate of the source node varies. 5) Normalized queue size: Queue size corresponds to the total number of the packets available in the queue at a time instance. 6) Congestion level: The node at which the congestion occurrence takes place is reflected using the congestion level. 7) Energy used: Energy used is sum of energies of links in the path. The energy in the transmission path network must be maximal for the improved performance. It depends on the energy needed to send and receive data between two nodes through a path and number of packets sent, measured in Joule/bits. 6.2.1 Evaluation based on throughput Figure 8 shows the analysis curve based on the throughput for the proposed rate optimization scheme. The analysis curve is plotted between the time (sec) and throughput. Throughput of the network must be increased for the improved performance. Figure 8 shows the analysis curve for the sensor network with 100 nodes and 10 s time instance. At each sec (time instance), the value of the throughput for the proposed and existing algorithms is plotted in the graph. The experimented results confirm that the performance of the proposed system is increased at every consecutive time instance. At the 5th time instance, throughput of the proposed system is 0.9574 whereas the throughputs of the existing systems are 0.8677 and 0.9086 for the CS and ACS

ACCEPTED MANUSCRIPT 18

AN US

CR IP T

system respectively. The analysis based on the throughput proves the proposed PSOGSA system to be more effective than the existing one.

Figure 8. Analysis based on throughput

6.2.2 Evaluation based on delay

AC

CE

PT

ED

M

The analysis curves based on the delay for the evaluation of the proposed system over the existing system are given in Figure 9. For a WSN, the delay for the transmission must be reduced for the optimal performance. In the network with 100 nodes, the delay analysis curve shown in Figure 9 confirms that the delay for the proposed system is gradually decreased over the time instance comparing with the existing systems. At the 10th instance, the delay of the proposed system is 0.0456 which is 0.0944 lower than the CS system and 0.0344 lower than the ACS. It shows the efficiency of the proposed method in terms of the delay. The proposed system in terms of the delay is efficient. The delay at every instance for the proposed system is lower than that of the existing systems, improving the performance of the WSN.

Figure 9. Analysis based on delay

ACCEPTED MANUSCRIPT 19

6.2.3 Evaluation based on normalized packet loss

M

AN US

CR IP T

The evaluation curve based on the packet loss is shown in Figure 10. The packet loss in the transmission must be reduced for the better performance of the network. Figure 10 shows the analysis curve relating the packet loss to time instance for the network with the 100 nodes. Every successive time instance increases the packet loss because of more incoming data packets. Even for the increased instance with increased packet arrival, the proposed system provides the minimal packet loss over the existing system. At 7th instance, the packet loss of the proposed system is 0.111762 whereas the existing system has the packet loss values as 0.191520 and 0.150916 for the CS and ACS system respectively. From the analysis curve, the performance of the proposed PSOGSA system provides minimal packet loss value compared to the existing system taken for the experimentation.

ED

Figure 10. Analysis based on normalized packet loss

6.2.4 Evaluation based on sending rate

AC

CE

PT

The sending rate of the data packets in the wireless sensor networks with 100 nodes is shown in Figure 11. In the network with 100 nodes, the sending rate at the time instance 7 for the PSOGSA is 0.030 whereas the CS and ACS systems have the sending rate of 0.0900.

Figure 11. Analysis based on sending rate

ACCEPTED MANUSCRIPT 20

6.2.5 Evaluation based on normalized queue size

M

AN US

CR IP T

The evaluation curve based on the queue size for the systems taken for the experimentation is shown in Figure 12. In this figure, for the network with 100 nodes, the analysis curve is plotted between the queue size and the time instance. For a WSN to provide the optimized performance, the queue size should be minimal. For the time instance 1, queue size attained by the PSOGSA is 0.0171 whereas queue size value of the systems CS taken for the experimental evaluation is 0.1. At the time instance 10, the queue size of the PSOGSA is less than 0.03 whereas the queue size attained by the CS is 0.1400 and ACS is 0.0800. For the successive time instance, the queue size of the proposed system is lower compared to the existing systems. The plot confirms the efficiency of the proposed system in terms of the queue size with the multiple networks.

Figure 12. Analysis based on normalized queue size

ED

6.2.6 Evaluation based on congestion level

AC

CE

PT

The analysis curve based on the congestion level for the evaluation of the proposed system is shown in Figure 13. This figure shows the congestion level analysis curve for the network with 100 nodes. In this figure, at the 10th instance, the congestion level of the proposed PSOGSA system is 0.1, at the same instance the existing system CS attains 0.3 which is 0.2 higher than that of the PSOGSA and ACS attains 0.2 congestion level value which is 0.1 higher than that of PSOGSA system. The value at the 10th instance proves the efficiency of the proposed system with the lower congestion level value compared to the other existing systems. The improvement in terms of congestion level is happened by properly adjusting the sending rate of the nodes.

Figure 13. Analysis based on congestion level

ACCEPTED MANUSCRIPT 21

6.2.7 Evaluation based on Energy used

AN US

CR IP T

The analysis curve based on the energy used for the evaluation of the proposed system is shown in Figure 14. This figure shows the analysis curve of energy used for the network with 100 nodes. In this figure, at the 10th instance, the energy used of the proposed PSOGSA system is 0.1, at the same instance the existing system CS attains 0.318233 which is higher than that of the PSOGSA and ACS attains 0.202087value which is also higher than that of PSOGSA system. The value at the 10th instance proves the efficiency of the proposed system with the lower energy used value compared to the other existing systems.

M

Figure 14. Analysis based on energy used

PT

6.2.8 Performance summary

ED

Based on the performance evaluation using the evaluation metrics such as congestion level, normalized queue size, normalized packet loss, throughput, and delay, the proposed PSOGSA system proves to be the best congestion control mechanism based on the rate optimization in the wireless sensor networks.

AC

CE

Table 5 shows the summary of the performance analysis. The values plotted in the Table 5are the parameters obtained after 10s of simulation time. From the table, we clearly proved that the proposed PSOGSA obtained the performance improvement in all the metrics. For example, throughput of the PSOGSA method is 0.5200 but the existing CS obtained only the value of 0.4200. Similarly, the delay is minimum for the proposed PSOGSA method in comparison to the existing methods. Reason for the improvement of the proposed mechanism is the adjustment of sending rate at every iteration. Table 5. Performance Analysis Summary

Time= 10s

CS

ACS

PSOGSA

Throughput

0.4200

0.4400

0.5200

Delay

0.1400

0.0800

0.0456

Normalized packet loss

0.5799

0.5599

0.4799

Sending rate

0.0961

0.0972

0.0971

Normalized queue size

0.1400

0.0800

0.0416

Congestion level

0.3000

0.2000

0.1000

Energy used

0.3182

0.2020

0.0999

ACCEPTED MANUSCRIPT 22

7. Conclusion

ED

M

AN US

CR IP T

We proposed a new congestion control mechanism for wireless sensor networks based on a hybrid multi-objective optimization PSOGSA. The ability of exploitation in PSO is combined with the ability of exploration in GSA to integrate the strength of both algorithms. PSOGSA is an efficient algorithm as the fitness or the quality of solutions is considered in each iteration. Agents having more attraction force, and close to the good solutions attract the other agents. When all agents are close to a good solution, they move slowly. PSOGSA uses a variable as memory to store the best solution found among all agents. Thus each agent can keep track of the best solution so far and move toward it. Thus, this concept has been used to reduce the congestion problem as the optimization algorithm avoids the congestion in the congested network by decreasing the average arrival rate of data from child nodes to the parent node, taking into consideration the energy of the node in its fitness function. The priority based transmission is enabled as the optimization approach regulates the arrival rate from every child node on the basis of its priority, output available bandwidth and energy of the child node. To mitigate the congestion, adjustment of rate to optimum value is used. Using PSOGSA, the optimization algorithm yields the new arrival rate from the child node which is less than that of the arrival rate of the last period. The optimization algorithm gives the fitness function based on the network congestion parameters based on which the control is provided. The selection of the new arrival rate for the child node is set using the selected fitness function value. Comparison of proposed protocol with existing protocol CS (Cuckoo Search) and ACS (Adaptive Cuckoo Search) algorithms was done. Simulation results prove the proposed mechanism is more efficient than existing CS and ACS algorithms in terms of performance metrics such as packet loss, end-to-end delay, Queue Size, throughput, Congestion level and Sending Rate. Improvement in Throughput is 5.50%, improvement in Delay is 44.47%, improvement in Packet Loss is 30.70%, improvement in Queue Size is 45.83%, improvement in Congestion Level is 45.83% and improvement in Sending Rate is 0.64%. Further studies of this research will investigate the load balancing problem and security issues in wireless sensor networks. Another direction can be seen from realistic scenarios. As the number of devices in smart cities and vehicular networks will be increased accordingly, our approach may be used to manage the load in vehicular networks in realistic roads or IoT for smart cities in future such as in [29-30].

1.

Antoniou (2007). Congestion control in wireless sensor networks. Master’s Thesis, University of Cyprus, Nicosia, Cyprus. Culler, Estrin D. & Srivastava M.(2004). Overview of sensor networks. IEEE Computer, 38(8), 41-49. Mirjalili, S., &Hashim, S. Z. M. (2010). A new hybrid PSOGSA algorithm for function optimization. In Computer and information application (ICCIA), 2010 international conference on (pp. 374-377). IEEE. Shi, Y., &Eberhart, R. (1998). A modified particle swarm optimizer. In Evolutionary Computation Proceedings, 1998. IEEE World Congress on Computational Intelligence (pp. 69-73). IEEE. Rashedi, E., Nezamabadi-Pour, H., &Saryazdi, S. (2009). GSA: a gravitational search algorithm. Information sciences, 179(13), 2232-2248. Rezaee, A. A., Yaghmaee, M. H., &Rahmani, A. M. (2014). Optimized congestion management protocol for healthcare wireless sensor networks. Wireless personal communications, 75(1), 11-34. Narawade, V., &Kolekar, U. D. (2016). ACSRO: Adaptive cuckoo search based rate adjustment for optimized congestion avoidance and control in wireless sensor networks. Alexandria Engineering Journal. DOI: 10.1016/j.aej.2016.10.005. Wilson, E. (1978). Sociobiology. The New Synthesis. Wan Y., Campbell A. T. & Eisenman S.B. (2003). CODA: congestion detection and avoidance in sensor networks. 1st ACM Conference on Embedded Networked Sensor Systems: (Sensys’03), pp 266-279.

AC

2. 3.

CE

References

4. 5. 6. 7.

8. 9.

PT

Acknowledgments: This work was carried out in Security and Computing laboratory, SC&SS, JNU, New Delhi, India and sponsored by the DST-PURSE grant.

ACCEPTED MANUSCRIPT 23

16. 17. 18. 19. 20.

21. 22. 23. 24.

25.

26.

AC

27.

CR IP T

15.

AN US

14.

M

13.

ED

12.

PT

11.

Ghaffari, A. (2015). Congestion control mechanisms in wireless sensor networks: A survey. Journal of network and computer applications, 52, 101-115. Ee T. &Bajcsy R. (2004). Congestion control and fairness for many-to-one routing in sensor networks. Proceedings of the 2nd International Conference on Embedded Networked Sensor Systems, Baltimore, Md, USA, November 2004, pp 148–161. Levis P., Patel N. & Culler D. (2004). Trickle: A self regulating algorithm for code propagation and maintenance in wireless sensor networks. Proceedings of USENIX/ACM Symposium on Network Systems Design and Implementation (NSDI), pp 15-28. Wang, Sohraby K., Lawrence V., Bo L., &Yueming H. (2006). Priority-based congestion control in wireless sensor networks. Proceedings of the IEEE International Conference on Sensor Networks, Ubiquitous, and Trustworthy Computing, June 2006, pp 22–31. Ahmad M. &Turgut D. (2008). Congestion avoidance and fairness in wireless sensor networks. Proceedings of IEEE Global Telecommunications Conference (GLOBECOM), pp 1-6. Sheu J. P. & Hu W. K. (2008). Hybrid congestion control protocol in wireless sensor networks. Proceedings of IEEE Vehicular Technology Conference, pp 213-217. Hussain, F. B., Cebi, Y., & Shah, G. A. (2009). A multievent congestion control protocol for wireless sensor networks. EURASIP Journal on Wireless Communications and Networking, 2008(1), 803271. Sergiou, C., Vassiliou, V., &Paphitis, A. (2014). Congestion control in Wireless Sensor Networks through dynamic alternative path selection. Computer Networks, 75, 226-238. Yang, X. S., & Deb, S. (2009). Cuckoo search via Lévy flights. In Nature & Biologically Inspired Computing, 2009. NaBIC 2009. World Congress on (pp. 210-214). IEEE. Antoniou, P., Pitsillides, A., Blackwell, T., Engelbrecht, A., & Michael, L. (2013). Congestion control in wireless sensor networks based on bird flocking behavior. Computer Networks, 57(5), 1167-1191. Majidi, A., &Mirvaziri, H. (2014). BDCC: Backpressure routing and dynamic prioritization for congestion control in WMSNs. International Journal of Computer Network and Information Security, 6(5), 29. Nguyen, T. T., Dao, T. K., Horng, M. F., & Shieh, C. S. (2016). An Energy-based Cluster Head Selection Algorithm to Support Long-lifetime in Wireless Sensor Networks. J. Netw. Intell, 1, 23-37. Taherkhani, N., & Pierre, S. (2016). Prioritizing and scheduling messages for congestion control in vehicular ad hoc networks. Computer Networks, 108, 15-28. Mahmood, M. A., Seah, W. K., & Welch, I. (2015). Reliability in wireless sensor networks: A survey and challenges ahead. Computer Networks, 79, 166-187. Kafi, M., Ben-Othman, J., Ouadjaout, A., Bagaa, M. and Badache, N. (2017). REFIACC: Reliable, efficient, fair and interference-aware congestion control protocol for wireless sensor networks. Computer Communications, 101, pp.1-11. Tam, N. T., Hai, D. T., Son, L.H., Vinh, L.T.(2018). Improving lifetime and network connections of 3D wireless sensor networks based on fuzzy clustering and particle swarm optimization. Wireless Networks, 1-14.Doi: 10.1007/s11276-016-1412-y. Hai, D. T., Son, L.H. & Le Vinh, T. (2017). Novel fuzzy clustering scheme for 3D wireless sensor networks. Applied Soft Computing, 54, 141-149. Son, L.H., Thong, P. H. (2017). Soft computing methods for WiMax Network Planning on 3D Geographical Information Systems. Journal of Computer and System Sciences, 83(1), 159-179. Hai, D. T., Tam, N. T., Son, L. H., &Vinh, L. T. (2016, December). A novel energy-balanced unequal fuzzy clustering algorithm for 3D wireless sensor networks. In Proceedings of the Seventh Symposium on Information and Communication Technology (pp. 180-186). ACM. Bazzi, A., Masini, B. M., Zanella, A., & Pasolini, G. (2015). IEEE 802.11 p for cellular offloading in vehicular sensor networks. Computer Communications, 60, 97-108. Silva, C. M., Masini, B. M., Ferrari, G., & Thibault, I. (2017). A survey on infrastructure-based vehicular networks. Mobile Information Systems, 2017. Sakhre, V., &Jain, S. (2014).Comparison of Hybrid PSOGSA & Cuckoo Search (CS) algorithm for Optimization of ETBE Reactive Distillation Vandana, Sanjeev Kang, J., Zhang,Y., & NathB. (2007) TARA: topology-aware resourceadaptation to alleviate congestion in sensor networks, IEEETrans. Parallel Distrib. Syst. 18 (7) 919–931.

CE

10.

28.

29. 30. 31. 32.

ACCEPTED MANUSCRIPT 24

33.

34.

AC

CE

PT

ED

M

AN US

CR IP T

35.

Magdy, A., EL-Ghandour, O. M., & Hamed, H. F. (2015) Performance Enhancement for Adaptive Beam-Forming Application Based Hybrid PSOGSA Algorithm, Journal of Electromagnetic Analysis and Applications 7(04) 126. Duman, S., Yorukeren, N., & Altas, I. H. (2015) A novel modified hybrid PSOGSA based on fuzzy logic for non-convex economic dispatch problem with valve-point effect. International Journal of Electrical Power & Energy Systems 64 121-135. Valdez, F., Melin, P., & Castillo, O. (2014) A survey on nature-inspired optimization algorithms with fuzzy logic for dynamic parameter adaptation. Expert systems with applications 41(14) 6459-6466.

ACCEPTED MANUSCRIPT 25

Appendix Detailed parameters of the PSOGSA as shown as follows. Congressionlevel.PSOGSA: Linear model Poly8: ans(x) = p1*x^8 + p2*x^7 + p3*x^6 + p4*x^5 + p5*x^4 + p6*x^3 + p7*x^2 + p8*x + p9

p1 =

-7.44e-07 (-3.801e-06, 2.313e-06)

p2 =

3.099e-05 (-0.0001036, 0.0001656)

p3 = -0.0005356 (-0.003016, 0.001945) 0.005002 (-0.01982, 0.02983)

p5 =

-0.0275 (-0.174, 0.119)

p6 =

0.09089 (-0.4252, 0.607)

p7 =

-0.1778 (-1.225, 0.869) 0.2013 (-0.8958, 1.298)

p9 =

-0.09133 (-0.5344, 0.3517)

M

p8 =

AN US

p4 =

Data.PacketLoss.PSOGSA Linear model Poly3:

ED

ans(x) = p1*x^3 + p2*x^2 + p3*x + p4

Coefficients (with 95% confidence bounds): 2.046e-20 (-1.232e-20, 5.324e-20)

PT

p1 =

p2 = -4.915e-19 (-1.537e-18, 5.541e-19) 0.003333 (0.003333, 0.003333)

CE

p3 =

0.006667 (0.006667, 0.006667)

AC

p4 =

CR IP T

Coefficients (with 95% confidence bounds):

Data.delay.PSOGSA Linear model Poly8: ans(x) = p1*x^8 + p2*x^7 + p3*x^6 + p4*x^5 + p5*x^4 + p6*x^3 + p7*x^2 + p8*x + p9 Coefficients (with 95% confidence bounds): p1 = -1.297e-06 (-7.075e-06, 4.481e-06) p2 =

6.021e-05 (-0.0001942, 0.0003146)

p3 =

-0.00117 (-0.005858, 0.003518)

ACCEPTED MANUSCRIPT 26

0.01233 (-0.03459, 0.05925)

p5 =

-0.07626 (-0.3531, 0.2005)

p6 =

0.2801 (-0.6954, 1.256)

p7 =

-0.5886 (-2.567, 1.39)

p8 =

0.6437 (-1.43, 2.717)

p9 =

-0.253 (-1.09, 0.5844)

Data.queuesize.PSOGSA Linear model Poly8: ans(x) = p1*x^8 + p2*x^7 + p3*x^6 + p4*x^5 + p5*x^4 + p6*x^3 + p7*x^2 + p8*x + p9 Coefficients (with 95% confidence bounds): 6.386e-07 (-4.57e-06, 5.847e-06)

p2 =

-2.716e-05 (-0.0002565, 0.0002022)

p3 =

0.0004842 (-0.003742, 0.00471)

p4 =

-0.004714 (-0.04701, 0.03758)

p5 =

0.02733 (-0.2222, 0.2769)

p6 =

-0.09596 (-0.9753, 0.7834)

p7 =

0.1952 (-1.588, 1.979)

p8 =

-0.1961 (-2.065, 1.673)

p9 =

0.09083 (-0.664, 0.8457)

PT

ED

M

AN US

p1 =

Data.SendingRate.PSOGSA

CE

Linear model Poly2:

ans(x) = p1*x^2 + p2*x + p3 Coefficients (with 95% confidence bounds):

AC

p1 = -5.681e-05 (-0.0001429, 2.927e-05) p2 = p3 =

CR IP T

p4 =

0.0006951 (-0.0002764, 0.001667) 0.09585 (0.09352, 0.09817)

Data.Throughput.PSOGSA Linear model Poly8: ans(x) = p1*x^8 + p2*x^7 + p3*x^6 + p4*x^5 + p5*x^4 + p6*x^3 + p7*x^2 + p8*x + p9 Coefficients (with 95% confidence bounds):

ACCEPTED MANUSCRIPT 27

p1 =

4.96e-07 (-3.848e-05, 3.948e-05)

p2 = -1.587e-05 (-0.001732, 0.0017) p3 =

0.0001806 (-0.03144, 0.0318)

p4 = -0.0007819 (-0.3173, 0.3158)

p6 =

0.009429 (-6.571, 6.59)

p7 =

-0.01925 (-13.37, 13.33)

p8 =

-0.00146 (-13.99, 13.99)

p9 =

1.012 (-4.637, 6.661)

CR IP T

p5 = -0.0001215 (-1.867, 1.867)

CE

PT

ED

M

AN US

Dr. Le Hoang Son obtained the PhD degree on Mathematics – Informatics at VNU University of Science, Vietnam National University (VNU). He has been working as a researcher and now Vice Director of the Center for High Performance Computing, VNU University of Science, Vietnam National University since 2007. His major field includes Soft Computing, Fuzzy Clustering, Recommender Systems, Geographic Information Systems (GIS) and Particle Swarm Optimization. He is a member of International Association of Computer Science and Information Technology (IACSIT), a member of Center for Applied Research in e-Health (eCARE), a member of Vietnam Society for Applications of Mathematics (Vietsam), Editorial Board of International Journal of Ambient Computing and Intelligence (IJACI, SCOPUS), Associate Editor of the International Journal of Engineering and Technology (IJET), and Associate Editor of Neutrosophic Sets and Systems (NSS). Dr. Son served as a reviewer for various international journals and conferences and gave a number of invited talks at many conferences.He has got 89 publications in prestigious journals and conferences including 41 SCI/SCIE, 2 SCOPUS and 1 ESCI papers and undertaken more than 20 major joint international and national research projects. He has published 2 books on mobile and GIS applications. So far, he has awarded “2014 VNU Research Award for Young Scientists”, “2015 VNU Annual Research Award” and “2015 Vietnamese Mathematical Award”.

AC

Dr. Karan Singh received the Engineering degree (Computer Science & Engineering) from Kamala Nehru Institute of Technology, Sultanpur, UP, India and the M.Tech (Computer Science & Engineering) from Motilal Nehru National Institute of Technology UP, India. He is Ph.D. (Computer Science & Engineering) from MNNIT Allahabad deemed university. He worked at Gautam Buddha University since Jan 2010. Currently, he is working with School of Computer & Systems Sciences, Jawaharlal Nehru University, New Delhi. His primary research interests are in Wireless Sensor Network, computer network security, Multicast communication and Software Define Network. He supervised 35 Master Degree students (M.Tech.) and 03 PhD. He is reviewer of IEEE transactions, Springer, Elsevier and Taylor and Francis journals. He is an Editorial Board Member of Journal of

ACCEPTED MANUSCRIPT 28

AN US

CR IP T

Communications and Network (CN), USA. He published more than 60 research papers in journal and good conference (including SCI/Scopus) and 02 Books. He organized of various workshops, Session, Conference and training. Dr. Singh worked as General Chair of international conference (Qshine) in year 2013 at Gautam Buddha University. Recently he is going to organize a workshop on “python”. He was nominated for Who’s who in World in year 2008. Dr. Singh has been joined as Professional member Association for Computing Machinery (ACM), New York, Computer Science Teachers Association (CSTA) U.S.A, Computer Society of India(CSI), Secunderabad, India, Cryptology Research Society of India (CRSI), Kolkata, India, Institute of Electrical and Electronics Engineers (IEEE), USA, International Association of Computer Science and Information Technology (IACSIT), Singapore, Institute for Computer Sciences, Social-Informatics and Telecommunications Engineering (ICST), America, International Association of Engineers (IAENG), Hong Kong, Association of Computer Electronics and Electrical Engineers (ACEEE), India, Internet Society(ISOC), USA and Academy & Industry Research Collaboration Center (AIRCC).

AC

CE

PT

ED

M

Karishma Singh received the B.Tech degree (Computer Science & Engineering) from Bhagat Phool Singh Mahila Vishwavidyalaya, Khanpur Kalan, Sonipat, Haryana, India and the M.Tech. (Computer Science & Technology) from School of Computer & Systems Sciences, Jawaharlal Nehru University, New Delhi, India. She is currently enrolled in Ph.D. (Computer Science & Technology) in School of Computer & Systems Sciences, Jawaharlal Nehru University, New Delhi, India. Her primary research interest is in Wireless Sensor Network and Optimization Techniques. She know the MatLab. She was organizing member workshop on Cyber and Network Security. She has joined as a member in Association for Computing Machinery (ACM), New York and Institute of Electrical and Electronics Engineers (IEEE), USA.