A Technique for Cluster Head Selection in Wireless Sensor Networks Using African Vultures Optimization Algorithm

INTRODUCTION: Wireless Sensor Network (WSN) has caught the interest of researchers due to the rising popularity of Internet of things(IOT) based smart products and services. In challenging environmental conditions, WSN employs a large number of nodes with limited battery power to sense and transmit data to the base station(BS). Direct data transmission to the BS uses a lot of energy in these circumstances. Selecting the CH in a clustered WSN is considered to be an NP-hard problem. OBJECTIVES: The objective of this work to provide an effective cluster head selection method that minimize the overall network energy consumption, improved throughput with the main goal of enhanced network lifetime. METHODS: In this work, a meta heuristic based cluster head selection technique is proposed that has shown an edge over the other state of the art techniques. Cluster compactness, intra-cluster distance, and residual energy are taken into account while choosing CH using multi-objective function. Once the CHs have been identified, data transfer from the CHs to the base station begins. The residual energy of the nodes is finally updated during the data transmission begins. RESULTS: An analysis of the results has been performed based on average energy consumption, total energy consumption, network lifetime and throughput using two different WSN scenarios. Also, a comparison of the performance has been made other techniques namely Artificial Bee Colony (ABC), Ant Colony Optimization (ACO), Atom Search Optimization (ASO), Gorilla Troop Optimization (GTO), Harmony Search (HS), Wild Horse Optimization (WHO), Particle Swarm Optimization (PSO), Firefly Algorithm (FA) and Biogeography Based Optimization (BBO). The findings show that AVOA's first node dies at round 1391 in Scenario-1 and round 1342 in Scenario-2 which is due to lower energy consumption by the sensor nodes thus increasing lifespan of the WSN network. CONCLUSION: As per the findings, the proposed technique outperforms ABC, ACO, ASO, GTO, HS, WHO, PSO, FA, and BBO in terms of performance evaluation parameters and boosting the reliability of networks over the other state of art techniques.

2 communication, and storage capabilities [4]. Other limitations of sensor nodes are node failure and network failure [5]. WSNs are hypersensitive and their lifetime is vulnerable to energy depletion of sensor nodes [6]. Optimal energy consumption in WSN is necessary to increase lifetime and performance. It can be achieved by performing clustering which decreases the energy consumption and increases the scalability of the network.
Clustering divides a network into equal or unequal clusters. Each cluster has a Cluster head (CH). CHs gather local data from cluster member sensor nodes, aggregate it, and transfer it to a distant base station (BS) directly or through other CHs [7,8]. The BS is linked to the Internet. Figure 1. represents the architecture of the WSN.

Figure 1. A WSN Architecture
In clustering, the selection of CHs is critical for improving the network durability since it affects member sensor nodes energy. As discussed in [9], CH-selection is an NP-hard optimization issue. As a result, the procedure of selecting the CHs is to be carried out with the utmost care.
The rest of the paper is divided into the following sections. The related work is presented in Section 2. Section 3 defines the network and energy model. Section 4 describes the proposed cluster head selection technique using AVOA. Section 5 discusses the nine meta heuristic techniques used for comparison with the proposed technique. Section 6 shows the simulation results and finally, section 7 concludes the paper.

Related Work
The essential contributions of researchers in the field of energy efficient clustering techniques have been discussed in this section. Many studies have been conducted by the researchers in the field of energy efficient cluster heads selection using conventional techniques as well as evolutionary techniques. These techniques are discussed here: In the year 2000, Heinzelman et al. [10] proposed a technique called Low-energy adaptive clustering hierarchy (LEACH) which was a probabilistic technique that randomly selects CH in each round. LEACH attained a large energy consumption while lengthening lifetime of the network compared to static clustering method. In the year 2002, Lindsey et al. [11] proposed PEGASIS which was termed as Power-Efficient Gathering In Sensor Information Systems refers to an approach that was chain-based. PEGASIS arranged the sensor nodes (SNs) so that they formed a chain, with each SN communicating only with its immediate neighbours. In the year 2011, Liu et al. [12] proposed Genetic Algorithm based LEACH in which cluster head was selected based on optimal value of cluster head probability using Genetic-Algorithm. It gave optimal probability of nodes which could be selected as cluster heads with minimum energy consumption.
In the year 2014, Sharawi et al. [13] proposed a technique based on Bat Swarm Optimization algorithm to select optimized cluster heads by minimizing the intra-cluster compactness with minimum distance between nodes in same cluster. In the year 2015, Gupta and Sharma [14] proposed a clustering algorithm based on modified Ant Colony Optimization using residual energy as a parameter. Comparative analysis was performed taking average energy of network, number of live nodes respect to number of rounds as performance evaluation metrics. In the year 2016, Rao et al. [15] proposed an energy efficient Particle Swarm Optimization (PSO) based cluster head selection protocol. Parameters such as intra-cluster distance, residual energy and sink distance of all the CHs were used in the fitness function. In the year 2017, Sengottuvelan and Prasath [16] proposed an improved Breeding Artificial Fish Swarm Algorithm for optimal selection of Cluster head in the network. The multi objective function was based on end to end delay and energy was formulated. In the same year, an energy-efficient clustering scheme called New Chemical Reaction Optimization, proposed by Rao and Banka [17], was based on a recent variable population-based chemoinspired approach (nCRO). It considerably increased the network's lifetime. However, CHs connect directly with the BS, which could be impractical in a large-scale network.
In the year 2018, Yogarajan and Revathi [18] presented Ant Lion Optimization for Clustering (ALOC), a technique to improve the energy efficiency of the network. The fitness function that was utilized in the ALOC took into account the residual energy, the number of nodes that were close to each node, the distance that separated the nodes from one another and the distance that separated each node from the BS. In the year 2019, Ahmad et al. [19] presented an approach for CHS based on an optimization technique called Artificial Bee Colony (ABC) method. The ABC's fitness function was evaluated on the basis of three parameters intra cluster distance, sink distance and residual energy. In the same year, Dattatraya and Rao [20] introduced a CH selection scheme using Glow worm Swarm Optimization (GSO) and Fruit Fly Optimization Algorithm (FFOA) to select the best CH in WSNs. Fitness Function was designed considering energy, distance, delay and QoS as important parameters. In the same year, an energy-efficient clustering scheme called New Chemical Reaction Optimization, proposed by Rao and Banka [17], was based on a recent variable population-based chemo-inspired approach (nCRO). In the year 2020, Prahadeeshwaran and Priscilla proposed a hybrid elephant V. Kusla  optimization algorithm called NIUS-HEHOA [21] to extend the lifespan of the network by selection energy balanced cluster heads. In the year 2021, Arunachalam et al. [22] introduced Squirrel Search Optimization-based Cluster Head Selection Technique (SSO-CHST) was presented for enhancing sensor network lifetime by using a gliding factor to determine cluster head selection during data aggregation and dissemination. The sensor node with the lowest fitness value was the cluster member. High-fitness sensor nodes were the possible cluster head.
From the literature, it has been deduced that choosing CHs for large-scale WSN is an NP-Hard problem, but that it can be addressed using optimization methods. Several methods for energy-aware sensor node clustering have been discussed in the literature as a result of the fact that energy efficiency plays an essential role in the WSN. It is generally agreed that one of the most essential aspects of a WSN is its ability to minimize the amount of energy that sensor nodes consumes. The researchers have focused the attention on the clustering and cluster head selection methods. Considering this as a motivation, in this work, we propose a cluster head selection technique using meta heuristic based African Vultures Optimization Algorithm (AVOA) to reduce the energy consumption of nodes in the network. The main contributions of this paper are given below: • A cluster head selection technique based on African Vultures Optimization Algorithm is proposed. • Various set of parameters are incorporated in this paper to evaluate a fitness function. • Comparison of proposed cluster head selection technique using AVOA with 09 state-of-the-art techniques in terms of average energy consumption, total energy consumption, network lifetime and throughput.

Preliminaries
In this section, network model and energy model, which are used in this paper, are discussed.

Network Model
The following are the characteristics of the WSN scenario considered in this paper. The sensors are distributed at random across the sensing field, and a node. The same methodology was used in the literature by [23]. As a result, no location-finding equipment, such as GPS, is required. After deployment, all sensor nodes are presumed to be stationary, and nodes can operate in cluster head and conventional sensor modes. Each node executes sensing on a regular basis and always has data to communicate to its CH or BS.

Energy Model
In this research, the classic Low-energy adaptive clustering hierarchy (LEACH) energy model [10,24] is used to calculate network energy consumption and exhausted energy for all network nodes. The same first order wireless communication model was used by [25]. Following assumptions have been made: • Initialize all the nodes with their attributes and establish their initial energies based on the first order radio energy model. • Node are transferring message with k bits via a distance d on symmetrical communication channels, and hence consume energy is expressed by equation (1) and equation (2). Based on the distance between the sender and receiver, the energy can be computed as follows: where ( , ) is the transmission energy consumption k bits data to a node, is the distance between sender and receiver nodes, is the energy dissipation per bit used to run the transmitter or receiver circuitry, _ is the amplifier parameter of transmission corresponding to the free space, _ is the amplifier parameter of transmission corresponding to the two ray model, 0 is the transmission distance threshold which is expressed by equation (3).
On other hand, the reception dissipation energy for message of bits for any node is expressed by the equation (4) due to running the receiver circuitry ( ).
( ) is the energy consumption in receiving k bits of data.

Proposed African Vulture Optimization Algorithm (AVOA) Based Cluster Head Selection Technique
In 2021, Abdollahzadeh et al. [26] proposed the AVOA meta-heuristic algorithm, which has since been used in a number of real-world engineering applications. Simulations and models based on the foraging behaviours and living habits of African vultures were used to develop the AVOA. Conversely, the best vulture right now is the strongest and most abundant vulture. In AOVA, vultures aim to be near the best and avoid the worst.
According to the aforementioned four norms, AOVA's problem-solving process may be broken down into five phases that mimic the actions of different vultures during the foraging phase.

Phase One: Identifying the best vulture in any group
After the formation of the initial population, the fitness of each solution is calculated, and the top and bottom performers are selected as vultures for the first and second groups, respectively. Populations are analysed comprehensively at each fitness iteration.
In equation (5), the chance that the chosen vultures will lead the others to one of the better solutions in each group is determined using 1 and 2 .
Phase Two: Vulture Hunger Rate They have high energy levels when they are full, so they can travel vast distances in quest of food, but when they're hungry, their energy levels are low and they can't walk as far as when they are full, so they become more aggressive. In order to model this phenomenon mathematically, equation (7) was applied in the process. For this reason, the rate at which vultures become full or hungry is taken into consideration while deciding whether or not to move from exploration to extraction. equation (7) has been used to describe the decline in the rate at which people become satiated.
1 is a random value between 0 and 1. When y is below 1, the vulture is starving; when it's 0, it's full.

Phase Three: Exploration
This stage examines AVOA's exploration phase. Vultures have excellent vision and can detect food and dying animals. Vultures meticulously inspect their surroundings and travel far to get food. In the AVOA, vultures can explore random areas using two distinct methodologies, and a parameter selects one. This option must be set between 0 and 1 before the search operation to determine which strategy is selected. A random integer between 0 and 1 is created when selecting a strategy in the exploration phase. Equation (9) is selected if the number is more or equal to the parameter . However, if number is smaller equation (11) is used. It is shown in equation (8).
A vulture's position vector in the following iteration is denoted by ( + 1), and its satiation rate in the present iteration is denoted by which is obtained using equation (7). ( ) is one of the best vultures selected by equation (5) in equation (10). Vultures move randomly in to guard food from other vultures. The formula = 2 × , where is a randomly generated number between 0 and 1, is used to create , which is then utilised as a coefficient vector to increase the random motion, which shifts with each iteration. The vulture's is the vector position. LB and UB show the variable bounds.
3 boosts randomness. If 3 is near to 1, similar solutions are distributed, adding a random motion to the LB.

Phase Four: Exploitation Stage-1
At this stage, the AVOA's efficiency stage is looked into. If the value | | is less than 1, the AVOA moves on to the exploitation phase. This phase also has two parts, each of EAI Endorsed Transactions on Scalable Information Systems 01 2023 -04 2023 | Volume 10 | Issue 3 | e9 A Technique for Cluster Head Selection in Wireless Sensor Networks Using African Vultures Optimization Algorithm 5 which uses a different strategy. Two parameters 2 and 3 , determine how likely it is that each strategy will be chosen in each internal phase. Parameter 2 selects first phase tactics, whereas 3 selects second phase strategies. Before running the search operation, both parameters must be set to 0 and 1.
| |between 1 and 0.5 begins the exploitation phase. During the first phase, two distinct rotational flight and siege-fight strategies are employed. 2 determines each strategy's choosing, which should be between 0 and 1 before searching operation performed. At the start of this phase, 2 is formed. If this number is larger than or equal to 2 , Siege-fight is implemented slowly. If the random number is less than 2 , rotating flight strategy is used. This procedure is shown in equation (12). where, ( ) is computed using equation (10), and is the vulture satiation rate, which is derived using equation (7). 4 is a number generated at random between 0 and 1.

Rotating Flight of Vultures
The model rotating flight pattern known as the spiral motion model is utilised frequently by vultures. With this strategy, a spiral equation is formed including every vulture and the top performer among the two. Equation (16) is used to describe the circular flight, and (17).
In (16) and (17), ( ) represents position vector of the two best vultures in the current iteration. 5 and 6 are random between 0 and 1. Equations (15) and (16) computes 1 and 2 . As a last step, the vultures' locations are revised using equation (17).

Phase Four: Exploitation Stage-2
The motions of the two vultures during the second stage of exploitation draw many vulture species to the food source, where siege and fierce competition for food occur. This phase is initiated when the | | value is less than 0.5. 3 , a value between (0,1) is generated during this stage. If 3 is more than or equal to 3 , numerous vulture species ought to congregate over the food supply. Alternately, the aggressive siege-fight strategy suggested in equation (18) is used if the generated value is lower than 3 .
Finally, all vultures are aggregated using equation (20), where 1 and 2 are from equation (19) and ( + 1) is the upcoming iteration's vulture vector. The top vultures in the first and second groups of the current iteration are respectively known as 1 ( ) and 2 ( ). The vector position of a vulture at any given moment is denoted by the symbol ( ), and the rate at which vultures become satisfied can be calculated using the equation (7). ( When | | is more than 0.5, the head vultures become starved, and they lack the strength necessary to contend with the other vultures. In order to model this motion, equation (21) is applied.
where ( ) reflects the vulture's distance from one of the two best vultures using equation (14).

Initialization and Solution Vector
To determine the ideal position of CH's in the network, a solution vector is employed to create an initial solution, which is processed using AVOA. The solution vector is composed of nodes (African Vultures) chosen at random as CHs from a total number of N sensor nodes. The position of each node is randomly assigned a node-ID from the total number of nodes i.e. N.
The dimensionality of each node agent is directly proportional to the number of CHs in the network. Assume that denotes the i th node in the network, and that the position of each node , is randomly assigned a node-ID between 1 to N. Solution encoding used for CH selection using AVOA optimization technique is illustrated in the figure below.

Fitness Function
The primary objective of proposed technique is to select energy efficient cluster heads using three parameters Intra Cluster Distance, Residual Energy, Cluster Compactness.
Since the three objectives are minimization problem in nature, so the ultimate fitness function F may be defined as a linear combination of 1 , 2 and 3 which is formulated as equation (22) : where, 1 , 2 3 are the weighted coefficients such that, 1 + 2 + 3 = 1 . The weight given are 1 =0.4, 2 = 0.3 and 3 = 0.3 respectively; 1 is Intra Cluster Distance, 2 is Residual Energy and 3 is Cluster Compactness.

Intra Cluster Distance:
The intra-cluster distance, or the distance that separates all of the sensor nodes in a cluster from the CH, is the first parameter that is evaluated for the objective function. Nodes depletes there energy when they communicate thus we must limit intra cluster distance. It means that CH may be selected to be closest to all cluster nodes. Thus, it is necessary to minimize the individual objective function 1 which is shown below in equation (23) : This function gives total distance for N number of sensor nodes for ℎ cluster heads. ℎ is the distance between cluster head and sensor nodes.

Residual Energy:
Residual energy of the cluster head is taken into consideration. The goal is to make the most use of all of the WSN's remaining energy of the sensor nodes. Since, the objective function is a minimization problem, so it is expressed as in equation (24) below: This function gives total residual energy of all the cluster heads.

Cluster Compactness:
Cluster Compactness represents measure of closer proximity among the normal nodes and CHs. It is third essential parameter that aids in generating clusters which comparably spend less energy. Cluster compactness of a node is determined as node degree divided by sum of distance to neighbour nodes. High cluster compactness value implies that the node can form compact clusters and lowest intra cluster communication cost is incurred. Hence, a node with high cluster compactness score has greater probability of getting elected as CH. It is expressed as in equation below : where, ( ) represents set of nodes at one hop distance from ℎ node. ND is node degree.

Evaluation of Fitness Function
In this stage, the formed fitness function is assessed for each location of the sensor nodes by inputting the values that are related with the solution vector (decision variable). The fitness function value for the solution vector is then represented in equation (26).
In this scenario, the fitness value of each sensor node depending on its position depicts how well they perform in terms of energy and minimal distances taken into account during the construction of the fitness function.
According to the solution vector, each sensor node in the network calculates its fitness value and arranges it in ascending order. The sensor node with the lowest fitness score will be selected as the network's cluster head. The cluster head in the following round may be chosen from among the remaining sensor nodes. This cluster head selection procedure, however, is reliant on the sensor nodes having enough energy because they must gather and transmit the data to the BS.

Meta-heuristic Techniques For Comparison
Meta heuristic algorithm, a method for solving optimization issues, begin by creating random response(s) and then advance to optimizing based on their operators and through modifying the generated random answers. In general, all meta heuristic algorithms use the similar approach to discover the optimal answer [27]. In the majority of these algorithms, the search process begins by producing one or more random solutions in a range of variables that is suitable. The primary generated solution in the population based algorithms is termed population, colony or group. and also each of solutions is named chromosome, particle, ant, and etc. [28] Then, utilizing operators and other methods of The new solution will also be selected from the pool of prior ones, and this process will carry on until the stop requirement is satisfied.
In this work, different meta heuristic techniques are used for comparative analysis with the proposed technique for CH selection i.e. based on African Vulture Optimization Algorithm. Meta-heuristic techniques used in this work are Artificial Bee Colony Optimization (ABC) [29], Ant Colony Optimization (ACO) [30], Atom Search Optimization (ASO) [31], Wild Horse Optimization (WHO) [32], Harmony Search Optimization (HS) [33], Gorilla Troops Optimization (GTO) [34], Firefly Algorithm (FA) [35], Particle Swarm Optimization (PSO) [36], Biogeography Based Optimization (BBO) [37]. Simulation parameters, average energy consumption, total energy consumption, network lifetime and throughput are used in this work for performance evaluation of the above mentioned techniques.

Result and Discussions
In this paper, we have proposed an energy-efficient CH selection technique based on African Vultures Optimization Algorithm (AVOA) and a fitness function considering intra-cluster distance, residual energy, and cluster compactness for the algorithm's energy efficiency. The algorithm compromises of two different scenarios namely WSN Scenario-1 for 100 nodes and WSN Scenario-2 for 200 nodes while considering same network area of 100 × 100 .
This section compares the performance of the techniques namely ABC, ACO, ASO, WHO, HS, GTO, FA, PSO and BBO based on the simulation parameters average energy consumption, total energy consumption, network lifetime and throughput. The simulation was run for 6000 rounds. However, 1000 rounds are used to evaluate the performance parameters of these techniques.

Simulation Parameters
Simulation parameters used for performance evaluation of the proposed cluster head selection technique are as follows: a) Average Energy consumption: It calculates the average energy difference between each node's original level and its remaining level [38].On a per-round basis, it is the amount of energy spent by a node to transmit the data in the WSN network. b) Total Energy Consumption: Network energy dissipation per round is a measure of how much power is used by the network's nodes [15]. c) Network Lifetime (FND): A WSN's life span is defined in this study as the number of rounds it goes through before its first node dies [39] . d) Throughput: The term 'throughput' refers to the total number of data packets that are successfully transferred to the sink [40]. Table 1. Parameter description

Total Energy Consumption Evaluation
As shown in Figure 4 and Figure 5, total energy consumption of the proposed technique with AVOA when compared with the rest of other techniques ABC [29], ACO [30], ASO [31], WHO [32], HS [33], GTO [34], FA [35], PSO [36], BBO [37] for 1000 rounds is the minimum. ABC technique consumes maximum energy while the proposed technique consumes lesser energy compared to rest of the techniques. Table 3 shows the comparative analysis in terms of total energy consumption. When the number of nodes in the network is doubled from 100 to 200 while the number of simulation rounds is kept the same, the overall energy consumption of the proposed technique AVOA is lower than that of ABC [29], ACO [30], ASO [31], WHO [32], HS [33], GTO [34], FA [35], PSO [36], BBO [37].The maximum energy consumption is by the technique with ABC.

Network Lifetime (FND)
First Node Die (FND), which specifies the first node that depletes its available energy throughout the network, is the metric that is used to determine the network lifespan in WSN. The results of comparing the network lifespan of the proposed technique AVOA with the rest of techniques with ABC [29], ACO [30], ASO [31], WHO [32], HS [33], GTO [34], FA [35], PSO [36], BBO [37] are shown in Figure 6. The AVOA has a FND of 1391 in Scenario-1 of the WSN and 1342 in Scenario-2 of the WSN, which is comparatively higher than the other techniques. It is shown that proposed technique has a higher network lifespan even if the nodes density increases from 100 to 200 nodes as compared with rest of the techniques.

Throughput
As shown in Table 4, throughput of the proposed technique with AVOA is maximum when compared with the rest of techniques with ABC [29], ACO [30], ASO [31], WHO [32], HS [33], GTO [34], FA [35], PSO [36], BBO [37]. It is observed that the proposed technique receives more data packets compare to the rest of the techniques. Also, when the node density is doubled, the proposed technique shows minimal decrease in throughput as compared with rest of the techniques.

Conclusion and Future Work
In conclusion, the lifespan of a network is an essential component of a WSN. Keeping track of how much energy is used is not an easy chore. We have proposed an energyefficient CH selection technique based on AVOA and a fitness function considering intra-cluster distance, residual energy, and cluster compactness for the algorithm's energy efficiency. Our proposed technique findings have been compared to well known techniques namely ABC, ACO, ASO, WHO, HS, GTO, FA, PSO, and BBO. This technique has been thoroughly tested with two different WSN scenarios. According to the experimental results, the suggested technique outperforms the traditional technique on the basis of average energy consumption, total energy consumption, network lifetime, and throughput. The results demonstrate that taking into consideration the aforementioned factors lead to improvements in both the average amount of energy used and the lifespan of the network. As future work, other factors effect in parallel with residual energy such as sensor range and transmission energy can be studied and evaluated. Our future research will be to create a routing algorithm utilizing a meta heuristic method.