An Improved Sunflower Optimization Algorithm for Cluster Head Selection in the Internet of Things

Due to the widespread of smart devices and services, the Internet of Things (IoT) has gained attention from researchers and is still in constant development. Many challenges face the IoT networks and need to be solved. Reducing energy consumption to increase the network lifetime is the main issue among these challenges. The clustering approach is one of the best solutions to solve this issue. Choosing the best Cluster Heads (CHs) can consume less energy in the IoT-WSN. Swarm Intelligence (SI) algorithms can help to solve complicated problems. In this paper, we propose a novel algorithm to select the best CHs in the IoT-WSN. The novel algorithm is called an Improved Sunflower Optimization Algorithm (ISFO). In the ISFO, we combine the Sunflower Optimization Algorithm (SFO) with the lèvy flight operator. Such invoking can balance the diversification and intensification processes of the proposed algorithm and avoid trapping in local minima. We compare the ISFO algorithm with six SI algorithms. The results of the proposed algorithm show that it can consume less energy than the other algorithms, also the number of nodes still alive for it is larger than alive nodes for the other algorithms. Hence, the ISFO algorithm proved its superiority in reducing the consumed energy and increasing the lifetime of the network.


I. INTRODUCTION
Internet of Things (IoT) has attracted great interest from researchers in recent years and is still in constant growth. IoT is a kind of network that involves a large number of physical devices also called '' Things '' and these devices are connected with the internet. The IoT network's device resources own limited processing, storage volume, bandwidth, and battery power capabilities [38], [39]. Since IoT can interpret the physical world to a digital world with useful information, it becomes an important concept in our lives. IoT is a network of objects that can sense, process, and transmit information through the use of sensors that could be a portion of a Wireless Sensor Network (WSN).
Wireless sensor networks (WSNs) play a significant role in the internet of things where they can supply services of The associate editor coordinating the review of this manuscript and approving it for publication was Amin Zehtabian .
sensing to the devices in the IoT through the sensor nodes [3]. Also in the area of network technology, the WSN is seen as of fundamental importance [11] and It had been employed in diverse applications [1], [10]. Figure 1 shows the structure of WSN.
IoT has various applications in several fields that can include education, health care, transport, smart cities, manufacturing, agriculture, military, environmental monitoring, smart grid et al. In these various applications, data collection is the function of the IoT devices. The amount of data collected is enormous and must be sent to the cloud for processing. The cloud may be available at a remote IoT network location. Due to the IoT limited resources, we cannot send the collected data directly to the cloud because it will make the nodes consume their energy rapidly and the network will die. To solve this issue, we must send the collected data to another base station (BS) located near the sensor nodes, and then these (BSs) forward the data to the cloud. It is necessary to manage the limited IoT network resources. The main IoT challenge between these resources is energy management [20], [24], [35], and [41]. Several energy management protocols have been implemented in IoT networks. However, in some cases, these protocols may fail to function. For this reason, clustering techniques can be a good solution.
Network clustering splits the network into some groups known as clusters. Each cluster contains several sensor nodes. Among these nodes, there is one node is chosen to be the cluster head (CH). The CH works as a local BS, which is responsible for collecting the data from other nodes and send it back to the remote BS (as the cloud in IoT). Chosen the optimal CHs will conserve more energy over the IoT network, so the network will run for a longer period. Figure 2 shows the structure of the IoT network, which contains an equal number of IoT devices and sensor nodes (SNs). where each SN is connecting to a specific IoT device. The function of each SN is to track and transfer the information to its IoT device. These devices and SNs are divided into clusters. In each cluster, there is one device is selected as a cluster head (CH) as A, B, C in Figure 2. Each CH is responsible for aggregating the data from other devices inside its cluster and then forwarding the aggregated data to the IoT BS (cloud).
Swarm intelligence (SI) algorithms have been utilized to solve complex problems [5]. In the last years, many SI algorithms have been applied in various fields and they have proved their superiority. In this paper, we proposed a distinct algorithm for the clustering process in IoT networks by merging the sunflower optimization (SFO) algorithm and the lèvy flight operator. The proposed algorithm is named ''an Improved Sunflower Optimization Algorithm (ISFO) and its function is to determine the optimal cluster heads. Consequently, more energy will be saved and the IoT network lifetime will increase.
The main contribution of this paper can be summarized as follows.
• a new SI algorithm is proposed to minimize the energy consumption of the nodes and increase their lifetime. • Invoking the lèvy flight operator in the proposed algorithm to avoid trapping in local minima.
• The proposed algorithm is proposed against six SI algorithms, the result of it can consume less energy than the other algorithms, also the number of nodes still alive for it is larger than alive nodes for the other algorithms. The remainder of the paper is arranged as follows. We survey the related work in Section II, we present the problem formulation in Section III. In Sections IV and V, we present the standard sunflower optimization algorithm and the proposed ISFO algorithm. We analyze the numerical results in Section VI. Finally, we summarize the paper in Section VIII.

II. LITERATURE REVIEW
Reddy and Babu in [30] use the Gravitational Search Algorithm (GSA) with Artificial Bee Colony algorithm (ABC) and consider some parameters to select the CHs efficiently. Consequently, save more energy and extend the network lifetime. Cui et al. [9] developed the LEACH protocol which is used for the big data sensing system in the IoT. They introduce a novel algorithm called WHCBA that involves the Bat Algorithm with a weighted harmonic centroid strategy. Then they merge the WHCBA algorithm with the LEACH protocol to improve the process of the cluster head selection and reduce the consumed energy as the centroid strategy improves the local search. Also in [31] Reddy and Babu proposed a novel way using the self-adaptive whale optimization algorithm (SAWOA) for improving the clustering protocols and reducing the energy consumption in the WSN based IoT networks.
In [12] Farman et al. have proposed an upgraded scheme for more CH proper selection in the IoT-based WSNs. This scheme depends on multi-criteria which includes remaining energy for a node, the distance among the node and its neighbors in the same zone(cluster), distance from the middle of the zone(cluster), how many times the node became zone(cluster) head. Finally, the node was merged into another zone(cluster) or remained in its zone. The proposed scheme helped the network save more energy and extend its lifetime. A hybrid model has been produced in [32] by Reddy and Babu for selecting the cluster heads in the WSN-IoT network. The model combines Ant Lion Optimization (ALO) and Moth Flame Optimization (MFO) algorithms. The aim was to conserve more energy and reduce the delay, load, distance, and temperature of IoT devices to make the network work for a long time.
A novel algorithm called '' R-LEACH '' has proposed by Behera et al. in [3] to achieve an efficient selection for cluster heads (CHs) in the network. The R-LEACH algorithm enhances the LEACH protocol to suit the IoT applications by taking some parameters when selecting the CHs. These parameters involve the energy of the node, at the beginning, its residual energy after each round, and the suitable number of CHs.
Another hybrid algorithm for CH selection has proposed by Janakiraman in [19]. The algorithm combines Artificial Bee Colony and Ant Colony Optimization algorithms. It enhances the residual energy rate for the WSN-based IoT network, decreases the number of dead nodes. As a result, the network's lifetime has extended.
A new hybrid clustering algorithm has introduced by Kannammal and Suresh in [21] to improve the level of energy and the working period for the network. To achieve this improvement, the algorithm uses the LEACH protocol with the Firefly algorithm.
In 2020 Iwendi et al. [18] focused on reducing the consumed energy for the SNs in the IoT network to prolong the network lifetime. They proposed a hybrid algorithm named (WOA-SA). This algorithm aims to choose the most suitable CHs in the IoT network to reduce the consumed energy. The WOA-SA algorithm utilized the whale optimization algorithm and the simulated annealing together. the CHs chosen depend on some metrics involving the sensor residual energy, temperature, load, number of alive SNs, and the cost function.
Rajesh et al. [28] minimized the energy consumption in the IoT network by implementing ''a multi-objective optimization routing algorithm'' called BFOA-R. The BFOA-R algorithm utilizes the foraging way of M. Xanthus and E. coli bacteria to extend the network time.
A novel scheme for data routing in IoT networks has been proposed by Nguyen in [27]. The objective of this scheme is to save more energy by using the compressed sensing (CS) technique when transferring various kinds of data gathered from the devices linked to the IoT networks.
Muhammad et al. [26] suggested a hybrid energy-efficient algorithm for the IoT-based WSNs by applying the scheduling approach. The algorithm is called HABCA-EST where the artificial bee colony (ABC) algorithm is combined with an efficient schedule transformation.
Al-Shalabi, Mohammed, et al. in [2] have used a genetic algorithm for producing a novel method called Optimal Multi-hop Path Finding Method (OMPFM). This method aims to reach the optimal path from the source (cluster head) to the destination (base station) for less energy consumption and maximizing lifetime in WSNs.
In [4], Bhatt et al. have utilized the cuckoo search algorithm to produce an enhanced algorithm for WSN. The objective is to choose the most suitable node as the cluster head, thus achieve efficient use of the network energy and extending its lifetime.
In [7], Chauhan et al. have offered a different technique for determining cluster heads in heterogeneous WSN. The technique is called ''DDMPEA-ANUM'' which refers to Diversity-Driven Multi-Parent Evolutionary Algorithm with Adaptive Non-Uniform Mutation. The goal of the ''DDMPEA-ANUM'' is to save more remaining energy for the sensor nodes and make the traveled distance less. Thus, the network will work for a long time.
Sarkar et al. in [34] have employed the firefly optimization algorithm (FF) and Grey Wolf Optimization algorithm to produce two models for ensuring the selection of the proper cluster heads among other sensors in WSNs. Thus, save energy and work the network for a long time. The models have been called ''Firefly Cyclic Randomization (FCR)'' and ''Firefly Cyclic Grey Wolf Optimization (FCGWO)''.
Fouad et al. [13] have employed the traditional GWO to achieve accurately locate the sink node. When the results have compared to one of the previous WSN protocols, they have shown higher accuracy.
Snasel et al. [36] achieved greater accuracy for allocating the sink node in WSNs by employing a novel strategy that relies on CSO in addition to the greedy algorithm. Another different algorithm for the WSNs was used by Fouad et al. [14] in order to determine the optimal sink node position, which helps improve the network performance. Chen and Li in [8] have been concerned with the problem of determining the best place for the sink node in WSNs by employing different strategies that focus on the network's energy and lifetime. In [6], Cao et al. propose CSSO, a novel algorithm for optimizing the placement of sensor nodes in heterogeneous WSNs. For achieving optimal coverage, chaos technique and SSO (social spider optimization) algorithm have been utilizing in this algorithm.
For solving the problem of network coverage and lifetime, multiple upgraded algorithms have been utilizing depend on the PSO algorithm by Ling et al. in [23]. Based on the simulation data, it has been demonstrating that the upgraded algorithms perform well and have a good deployment effect. An advanced fish swarm optimization algorithm called AIFS has been employing by Qin and Xu in [37]. The objective of AIFS is to free the nodes from the local optimum and ensure that the monitoring area is effectively covered. Rajpoot and Dwivedi [29] have suggested a MADM strategy for choosing the fittest node placement with achieving the most region coverage, more connectivity, and lowest expense among a variety of choices.
We can summarize the current issues and problems in the IoT and WSN in Table 2 and the list of the used parameters abbreviations is shown in Table 1 as follows.

III. PROBLEM DEFINITION
As energy is considered one of the most valuable resources in the IoT, our work aims to achieve effective management of energy utilization, which will lead to an expansion in the lifespan of the IoT network. We can do this by applying the clustering approach, where the optimal choice for the CHs in the IoT network will lead to consuming less energy. Consequently, The network will operate for a longer time. In our ISFO algorithm, clustering happens in two steps: Selection of CHs then formation of the clusters. We will explain these two steps in the following subsections.

A. SELECTION OF CLUSTER HEADS
The selection process of the CHs in the proposed ISFO algorithm takes place by applying a different fitness function, which based on a number of parameters as demonstrated in the following: • Average distance between CHs and SNs. It refers to the summation of the distances between each CH j(CH j ) and all SNs s i . Then we calculate their average as shown in Equation 1 1 where m refers to the number of CHs and N is the total number of SNs. When each node transmits data to its CH, it consumes some energy. For that, we have to select CHs near all the remaining SNs to reduce the consumed energy.
• Average distance between CHs and BS. It points to the distance between each CH j(CH j ) and the BS (BS) divided by the number of CHs (m) as written in Each CH collects the data from its SNs, and starts to forward these data to the BS. So it is better to pick the CHs that close to the BS. Consequently, we can merge Equation 1 and Equation 2 in Equation 3 (named it f distance ) because we want to minimize the distances between cluster heads and nodes and the distance between the base station and each cluster head.
Total energy for CHs. This parameter refers to the sum of the current energy for all the picked CHs. Our purpose is to maximize this sum to pick the optimal CHs. In another word, We aim to minimize the inverse of this sum as shown in equation 4 (named it f energy ). Because each node expenses some energy when transmitting the data.
It is important to pick the CHs from the nodes that own more energy than other nodes.
E ( CH j ) is the current value of energy for a cluster head j where (1 ≤ j ≤ m). From the previous two functions ''f distance and ''f energy we can form the fitness function by merging these functions into one function called ''F fitness as shown in Equation 5 Min Where R is the maximum range of communication for each SN s i , R max is the maximum range of communication for each CH, SNs is the group of all the sensor nodes, C is the group of all CHs, C = {CH 1 , CH 2 , . . . , CH m }, TH E refers to ''threshold energy'' to be a CH, α is a control parameter.
For selecting the optimal CHs, we aim to minimize the value of the fitness function in Equation 5. The smaller the fitness value, the best CH position we have.

B. FORMATION OF CLUSTERS
The formation of the clusters happens after the first step has been finished. Cluster formation takes place by using a weight function called '' WeightF '' this function depends on the following parameters: • Residual energy for the CH. For a SN s i , It should combine to a CH j(CH j ) that owns more residual energy than   other CHs in its communication range. Consequently, E residual (CH j ) is pointing to the residual energy for a CH j.
• Distance between the SN and CH. For a sensor node s i , it should combine to the closest CH j(CH j ) in its communication range. Where this will help in consuming less energy. Consequently, • Distance between the CH and BS. CHs are responsible for receiving the data from the SNs, and forwarding them to the BS. For this reason, a SN s i should combine to a CH that is closer to the BS than other CHs in its communication range.
• Degree of the CH node. For a SN s i , it should combine to a CH j(CH j ) that owns the least node degree in its communication range. For this,
Consequently, the final weight function for cluster formation as in Equation 11 WeightF(s i , where, C refers to a constant and its value is equal to 1.
To form the clusters, each SN uses Equation 11 to calculate its ''WeightF and then it must combine to a CH that owns the largest weight value.

IV. SUNFLOWER OPTIMIZATION ALGORITHM (SFO)
The Sunflower optimization algorithm (SFO) is a natural inspired algorithm proposed by G.F. Gomes et al. in 2019 [15]. The algorithm mimics the pollination process between the nearest two sunflowers during the movement toward the sun. In the next subsection, we highlight the characteristic of the SFO algorithm and the main processes of it.

A. THE NATURAL BEHAVIORS
Every morning, the sunflowers move toward the sun and the pollination process can happen between the nearest two sunflowers X i and X i+1 . Each sunflower absorbs the radiation from the sun. The amount of aggregated radiation for each sunflower depends on its position to the sun. The longer distance between the sunflowers and the sun, the lower amount of the received radiation (heat) from it. The amount of the received heat for each sunflower from the sun can be represented as shown in Equation 12 where Q i is the amount of the received heat, W is the sun power and c is the distance between the best solution (sun) X * and the sunflower X i .

B. THE SUNFLOWER ORIENTATION ADJUSTMENT PROCESS
The orientation vector for each sunflower is calculated as shown in Equation 13. where X * is the global best solution, X i is solution i, and N is the population size. The sunflower orientation adjustment process is shown in Figure 3.

C. STEP SIZE OF THE SUNFLOWERS TOWARD THE SUN
The step size of each sunflower X i towards the sun is calculated as shown in Equation 14.
where α represents the sunflower's inertial displacement, P i ( X i + X i−1 is the probability of the pollination between the closest two sunflowers X i and X i+1 . The closer sunflowers to the sun take a smaller step to refine their positions (exploitation process) while the more distance sunflowers move randomly (exploration process). The step size for all sunflowers is restricted to the maximum step size d max to avoid skipping from the boundary for each solution. The maximum step size for each sunflower is calculated as shown in Equation 15.
where X max is the upper bound, X min is the lower bound, and N is the population size.

D. FERTILIZATION PROCESS
The best sunflowers will fertilize around the sun to generate new individuals. The fertilization process for each sunflower can be represented as shown in Equation 16 where X i+1 is new generated sunflower.

E. THE SFO ALGORITHM
The main processes of the SFO algorithm are shown in Figure 4.

V. AN IMPROVED SUNFLOWER OPTIMIZATION ALGORITHM (ISFO)
The standard SFO algorithm like the other SI algorithms suffers from slow convergence and it can be trapped easily in local minima. To increase the diversity search in the standard sunflower optimization algorithm (SFO), we invoke the lèvy flight operator on it to produce a new version of the SFO algorithm, which is called an Improved Sunflower Optimization algorithm (ISFO). The lèvy flight operator is a random walk method that can help the SFO escaping from trapping in the local minima and avoid premature convergence. The main steps of the proposed ISFO algorithm are the same as the standard SFO algorithm except that we replace the step size in Equation 14 with the lèvy flight operator as shown in the following Equation.
X i+1 = X i + levy(ν) × s i (17) where Levy (ν) is the lèvy distribution. The structure of the proposed ISFO algorithm is presented in Algorithm 1.
We can summarize the steps of the proposed ISFO algorithm 1 as follow. i ). 5: The overall best solution is assigned X * . 6: repeat 7: All solutions adjust their orientation toward the sun (best solution)X * as shown in Equation 13. The new solutions are accepted if their fitness are better than the current solutions. 12: Set t = t + 1. 13: until (t > max itr ). 14: The overall best solution is presented. the population size N , the rate of the pollination P and the maximum iteration number max itr .
• Iteration counter initialization. The counter for the initialization is initialized t := 0. • Iteration Counter increment. The counter for the iteration number is increased t = t + 1.
• Termination criteria satisfaction. The processes are repeated until reaching to the maximum number of iterations max itr . • The global best solution. The overall best solution is presented.

VI. EXPERIMENT RESULTS
We implemented the proposed ISFO algorithm by utilizing MATLAB R2019b and this on an Intel Core i5 processor 2.30 GHz, 8 GB RAM working on the operating system Windows 10 Pro (64-bit).

A. PARAMETER SETTING
In Table 3, we demonstrate the IoT network parameters.
Where these parameter values have been utilized by Heinzelman et al. previously as in [17].
Where ND is the (size-diameter-target area-sensing) field of the IoT network and it equals (200m × 200m). BS represents the position of the IoT base station and we have three cases at the points (100,100), (200,200), and (300,300). SN is the overall number of sensor nodes that are used in our IoT network and is equal to 300 nodes. CHs is the number of the selected CHs and it represents 10% of the total number for the SNs (i.e 300 × 10% = 30 CHs). E o refers to the initial energy amount for each SN and is equal to 2 Joules. E TX and E RX indicate the amount of energy dissipated for sending and receiving a bit of data respectively and are equal to 50 nJ/bit. fs and mp denote the amount of energy dissipated by the transmitting amplifier for both free space and multi-path models are equal to (10 PJ/bit/m2), (0.0013 PJ/bit/m4) respectively. R is the sensing radius that means the maximum range of communication for each SN and is equal to 100m. d o = 30 m and it denotes the distance of threshold transmission. K means that the data package size = 4000 bits. Finally, EDA is the energy required to aggregate the data and is equal to 5 nJ/bit.
In Table 4, we demonstrate the values of the proposed ISFO algorithm parameters. Where the parameter values have been taken from the original paper of the sunflower optimization algorithm [15].
From Table 4, N refers to the number of populations (sunflowers) that equal to 30. D = 30 is the dimension of our problem that means the number of the selected CHs. P, m, and s indicate the rate of pollination, mortality, and survival  for the sunflowers respectively. α is a control parameter where the energy and distance parameters are controlled by α. Max itr refers to the maximum number of iterations that equals 5000 iterations.

B. EXPERIMENT SETTING
We have distributed 300 nodes randomly where the network dimension 200 × 200m 2 . There are three different scenarios of the BS location in our network. The network dimension is the same for all cases. The BS position is in the center of the network at (100,100) in the first case as shown in Figure 5, while the BS position is in the top right corner of the network at (200,200) in the second case as shown in Figure 6. Finally, the BS position is outside the network domain (300,300) as shown in Figure 7.
The following subsections show the efficiency of invoking the lèvy flight operator in the proposed ISFO algorithm then we compare its performance with other algorithms. We ran all algorithms 10 times.

C. TIME COMPLEXITY OF THE PROPOSED ISFO ALGORITHM
The time complicity of the ISFO algorithm can be computed as follow.
• Initial population. The time complexity of the initial population is O (N × D), Where N is the population size and D is the problem dimension.

D. THE PERFORMANCE OF THE PROPOSED ISFO ALGORITHM
From Equations 13 to 16, we can see that each solution in the population can update its position based on the position of the overall best solution. However, if the overall best solution gets in local minima, the rest of the population will trap in this region. Based on this issue, we invoke the lèvy flight operator in the proposed ISFO algorithm to increase the diversity of it. To show the diversity of the proposed ISFO algorithm, we show the convergence curves of the standard SFO and the proposed ISFO as shown in Figures 8,9,10,11,12,13. Figures 8,9,10 show the relation between the number of round (iteration) versus the number operating nodes. The solid line represents the results of the ISFO algorithm, VOLUME 9, 2021   while the dotted line represents the results of the standard SFO algorithm.
At the beginning of the search, both algorithms have the same diversity during the search, however, after a number of iterations, the solutions in the standard SFO algorithm get in local minima, while the proposed ISFO avoids trapping in local minima. Figures 8, 9, 10 demonstrate the number of nodes that still alive (operating nodes) after 5000 rounds and the BS at the  points (100,100), (200,200), and (300,300) respectively when using the SFO algorithm and the proposed (ISFO) algorithm.
From Figures 8, 9, 10, we can see that the number of nodes still alive for the ISFO algorithm is larger than alive nodes for the SFO algorithm.
We apply another test to verify the diversity of the proposed algorithm by plotting the relation between the energy consumption versus the round (iterations) as shown in Figures 11, 12, 13. In Figures 11, 12, 13, the solid line represents the results of the proposed ISFO algorithm while the dotted line represents the results of the standard SFO algorithm.
In Figures 11, 12, 13, at the beginning of the search, the convergence curves of both algorithms show that they have the same diversity effect. However, when the number of iterations increased, the diversity of the proposed ISFO increased while it decreased in the SFO algorithm. Figures 11, 12, 13 show the total consumed energy for the SFO and the proposed ISFO algorithms after 5000 rounds. The BS at the position (100,100), (200,200), and (300,300) respectively.
In Figures 11, 12, 13, it is clear that the consumed energy for the proposed ISFO algorithm is less than the consumed energy for the SFO algorithm. We can conclude from the previous figures, that invoking the lèvy flight operator in the proposed ISFO, can increase its diversity and help it to avoid trapping in local minima. We compare the ISFO algorithm with the LEACH Protocol [17] and other SI to prove its superiority. The algorithms include Grey Wolf Optimizer (GWO) [25], Particle Swarm Optimization (PSO) [22], Harris Hawks Optimization (HHO) [16], Artificial ecosystem-based optimization (AEO) [40], Grasshopper Optimization Algorithm (GOA) [33] and Sunflower Optimization Algorithm (SFO) [15]. The results of the comparison are plotted in Figures 14, 15 and 16 according to the number of nodes that still alive (operating nodes) after 5000 rounds and the BS at the positions (100,100), (200,200), and (300,300) respectively.
The GWO, PSO, HHO, AEO, GOA, SFO are swarm intelligence and naturally inspired algorithms, which suffer from the slow convergence, and their solutions update their position based on the position of the overall best solution. If the best solution gets in the local optima, they follow it and can be trapped in that local minima. Figures 14, 15 and 16 show the relation between the operation node and the round (iterations) by plotting the convergence curves of the proposed ISFO and the other algorithms.
From the Figures 14, 15 and 16, we can note that all algorithms have the same operation nodes, however, increasing the number of iterations reduces the number of operation nodes of all algorithms except the proposed ISFO algorithm because its diversity and ability to avoid trapping in local optima.
The diversity effect of all algorithms is verified also in Figures 17, 18 and 19 by showing the total energy consumption after 5000 rounds. All algorithms consume the same energy at the beginning of the search while increasing the number of iterations increases their consuming energy rapidly except the proposed ISFO algorithm has a lower consumed energy.   The numerical results of all algorithms concerning the average (Avg), minimum (Min), maximum (Max), standard deviation (Std) energy consumption after 5000 rounds are reported in Tables 6, 7  From the results shown in Tables 6, 7 and 8, we can see that the consumed energy by the ISFO algorithm is less than the remaining algorithms, and it is superior to the other VOLUME 9, 2021    algorithms and can help the network to operate a longer time than others.

VII. CONFLICT OF INTEREST
In the present work, we have not used any material from previously published. So we have no conflict of interest.

VIII. CONCLUSION AND FUTURE WORK
The network lifetime is the main issue that faces the IoT-WSN. To overcome this issue, we proposed a new algorithm to select the CHs in the IoT-WSN, which is an NP-hard problem. The proposed algorithm is called an Improved Sunflower Optimization Algorithm (ISFO). The ISFO algorithm is a hybrid algorithm between the standard SFO algorithm and the lèvy flight operator. The lèvy flight operator can increase the diversity search of the ISFO algorithm and help it to escape from local minima. The ISFO is compared with six SI algorithms at different BS positions. The results of the ISFO algorithm show that it can increase the network lifetime compared to the other algorithms. Our future work is to test the ISFO algorithm in a dynamic BS position and deal with real-time data and combining more operators into the ISFO algorithm to improve its performance.