Green Communication in Internet of Things: A Hybrid Bio-Inspired Intelligent Approach

Clustering is a promising technique for optimizing energy consumption in sensor-enabled Internet of Things (IoT) networks. Uneven distribution of cluster heads (CHs) across the network, repeatedly choosing the same IoT nodes as CHs and identifying cluster heads in the communication range of other CHs are the major problems leading to higher energy consumption in IoT networks. In this paper, using fuzzy logic, bio-inspired chicken swarm optimization (CSO) and a genetic algorithm, an optimal cluster formation is presented as a Hybrid Intelligent Optimization Algorithm (HIOA) to minimize overall energy consumption in an IoT network. In HIOA, the key idea for formation of IoT nodes as clusters depends on finding chromosomes having a minimum value fitness function with relevant network parameters. The fitness function includes minimization of inter- and intra-cluster distance to reduce the interface and minimum energy consumption over communication per round. The hierarchical order classification of CSO utilizes the crossover and mutation operation of the genetic approach to increase the population diversity that ultimately solves the uneven distribution of CHs and turnout to be balanced network load. The proposed HIOA algorithm is simulated over MATLAB2019A and its performance over CSO parameters is analyzed, and it is found that the best fitness value of the proposed algorithm HIOA is obtained though setting up the parameters popsize=60, number of rooster Nr=0.3, number of hen’s Nh=0.6 and swarm updating frequency θ=10. Further, comparative results proved that HIOA is more effective than traditional bio-inspired algorithms in terms of node death percentage, average residual energy and network lifetime by 12%, 19% and 23%.


Introduction
Over the years, revolutionary development in IoT devices has opened the paradigm for dynamic sensing technology that provides seamless communication over the Internet [1]. Wireless sensor networks (WSNs) are prominently used for the collection of data and communication over the fifth generation (5G) and beyond 5G IoT network envision as sixth generation (6G) technology [2]. Moreover, the combination of IoT networks with WSNs has many potentials in various applications, such as precision agriculture, intelligent transport systems, health care, smart cities, military, environment and habitat monitoring, environment anomalies and human intrusion detection [3,4]. However, with all these remarkable properties, the fallout in the unbalanced energy consumption and lower lifetime

•
The system model includes cluster-based IoT architecture with aid the feature of cloud network, where energy consumption of the cluster network is evaluated. • An energy optimization problem is formulated in terms of a fitness function that minimizes the intra-and inter-cluster distance in the IoT network.

•
We present HIOA to generate an optimal set of CHs to minimize the overall energy consumption. This employs FL for the creation of the initial population. Further, CSGO divided the IoT node into a hierarchal structure to increase the population diversity that optimizes the formulated fitness function using crossover and mutation. • Finally, extensive simulation over different CSGO parameters and comparison with the state-of-the-art algorithms has been performed for critical performance evaluation.
The rest of the paper is divided into the following sections. Section 2 shows the related literature. Section 3 describes the presented system model, problem formulation and proposed algorithm in detail. In Section 4, the simulation results and analysis are discussed. The conclusion of the work and future perspective are presented in Section 5.

Related Works
As mentioned above, recent studies [13][14][15][16][17][21][22][23][24] through simulation have observed that artificial intelligence and precisely bio-inspired techniques are preferable to traditional probabilistic and deterministic approaches subject to optimizing the energy consumption in IoT networks. In [17,21], the authors proved the proposed algorithm based on CSO obtains excellent performance over traditional approaches such as PSO, FL and GA for robust beam-forming approach, whereas authors in papers [24,25] CSO-based clustering routing protocol plays a significant role in reducing energy consumption over integration with bio-inspired approaches.
In paper [26], a Fuzzy-based routing protocol (FRP-LEACH) is proposed to enhance the traditional Low Energy Adaptive Clustering Hierarchy (LEACH) protocol that forms the clusters in an energy efficient manner. The proposed FRP-LEACH works over crosslayers, and the authors claim that cloud-based services such as the proposed algorithm protect medical staff and patients from the ongoing COVID-19 pandemic. However, the proposed algorithm is restricted to a lower dataset as the COVID-19 dataset increases, then fuzzy logic fails to adapt the changes delay in transferring the packet and node death ratio increases according to data growth rate. The authors in paper [27] enhanced the performance of LEACH using a fuzzy logic inference system and renamed LEACH-FL. However, both approaches carry forward the clustering process efficiently by considering the fuzzy logic inputs as closeness to the base station and residual energy but left out the node density parameter. This cause's optimum number of CHs generated through approaches to cover the entire network was not ensured.
In [28], authors have proposed EC-PSO models based on a standard PSO approach to optimize the energy consumption of the network to overcome the hotspot problem. When some of the sensor nodes may be left out from the coverage area of CHs and live spontaneously, they are called isolated nodes. These nodes continuously search for CHs and forces to communicate base station directly to exhaust more energy. However, the authors completely ignore the node distance from the base station, which causes it to affect the fitness function tremendously and ultimately consume uneven energy in each time slot. In [29], the authors proposed an enhanced version of a cluster-based genetic algorithm referred to as CRCGA by coding the fitness function to minimize the energy consumption and load balancing of the network. This proposed algorithm mainly considers three factors: formation of clusters, finding the best route and then maintaining the clusters appropriately in each time slot. Further, adaptive round-trip time is considered over the traditional TDMA schedule to further improve network performance. However, CRCGA outperforms the traditional algorithm but fails to consider the scenarios of different sink node positions. In the paper [30], authors have reduced the localization error in WSN using the CSO technique, and further statistical analysis compared to benchmarks optimization technique PSO and GA reveals that CSO provides an upper hand to robustness, precision and performance in terms of convergence speed over the IoT network. To boost the performance of CSO, a cuckoo search is integrated called CSCSO [31] to find the optimal route for data transfer between node and base station. An enhanced version of LEACH using CSO is presented as LEACH-PSO [32] to form optimal clusters and routing paths. Whereas in the paper [33], authors have addressed the problem of balance between power supply and demand using the CSO technique in the residential area and tertiary industry. Further simulation proved that the improved CSO algorithm outperforms in interruptible load scheduling over peak demand in real time over the GA and PSO algorithms. The above-mentioned algorithms missed the multi-hop routing scenario that extends to inter-clustering routing and intraclustering routing exchange a greater number of packets that maximize the overall energy consumption of the network.
From the above comparative study of Table 1, it is clearly demanding that bio-inspired CSO and genetic candidate are the best suited approach to handle the uneven clusters formation and efficiently optimize the overall energy consumption. In this paper, clustering is divided into two phases. (i) Fuzzy logic is used to produce tentative CHs. (ii) Thereafter, CSO with GA is used to produce the final optimal number of CHs at each round considering the minimization of total energy consumption per round focuses on enhancing the IoT network lifetime.

Network Model
In the energy-constraint IoT network, all the smart IoT nodes are deployed in the environment randomly. These sensor nodes form clusters as a Fuzzy logic-based chicken swarm genetic optimization (HIOA) algorithm. Only CH is able to send aggregated data to nearby edge nodes. Finally, edge node transferred data to the cloud network for data storage that can be used for data analytics with the following assumptions about IoT nodes below. All the smart IoT sensor nodes are homogenous in nature in terms of energy and computation capabilities (such as data collection, transmission and data aggregation). All these wireless IoT nodes periodically observe the environment and send data to their nearby CH. Initially, all the IoT nodes start with an equal quantity of energy stored in the battery. Using the received signal strength indicator (RSSI), IoT nodes calculate their distance from edge nodes by finding their co-ordinate (X l , Y l ). An equal amount of energy is consumed by the symmetrical channel in packet transfer in either direction from A to B or B to A. In addition, IoT nodes can adjust their transmission power for packet transfer according to distance either from the edge node or CH. The operational period is divided into two phases: the setup phase and the steady state phase. A crystal-clear observation of a single transmission period is further explained in Section 3.3.4 Transmission Procedure of HIOA Algorithm.

Energy Model
The IoT nodes set their states between transmitter and receiver before the start of the transmission period. The total energy (E S (y, z)) consumed to send y bit of packet is twofold (a) energy (E T ) exhaust in transceiver circuit to process the packet (b) energy (E A (y, z)) consumption takes place to amplify the y bit of packet over z distance either in free-space energy (∂ f s ) or multipath energy (∂ mp ) corresponds to threshold distance z 0 similar to the first order radio model [3] as follows: Additionally, energy (E r ) consumed by IoT node to receive y bit of packet as follows: We consider N IoT nodes are evenly distributed into W clusters such that N/W nodes are allocated to each cluster. Further, CHs use the TDMA slotted period to get data from their member IoT nodes, where IoT nodes are able to transmit the sensed data to their CH only in wakeup mode after that switch to sleep mode. The total energy (E P ) exhaust by each CH in a single transmission period is the energy payoff in (a) (E g ) gathering the data from its member's node and (b) (E e ) transmitted the received data to the edge node as aggregated data following the multipass model can be written as: where, E g = yE T N w − 1 , and E e = yE ay N W + yE T + y∂ mp z 4 pe is consumed energy in the multipass model. E ay represents the data aggregation energy and y ze denote the distance between CH and edge node. The energy consumed (E c ) by child IoT nodes to transfer gathered data to their parent CH uses a free space model over average distance (z cp = (1/2π) R 2 /w , R is diameter of network) between parent and child node in one round as follows: Overall, each cluster exhausts the total energy is the summation of energy consumed by parent CH and its child IoT nodes during one round as follows:

Hybrid Intelligent Energy Optimization
The primary objective of the proposed HIOA approach is to generate an optimal set of CHs to minimize the overall energy consumption and prolong IoT network lifetime, as shown in Figure 1. It consists of two optimization phases. First, a tentative set of CHs is selected through fuzzy logic inference system (FLIS) optimization. In the second optimization phase, the chicken swarm genetic optimization (CSGO) algorithm uses the output of the FLIS labeled as input and treated as the first initial population. Further, the

Adapted Fuzzy Logic Interference System for Green Communication
In the first optimization phase, a tentative number of CHs is selected based on three parameters in an IoT network: residual energy ( ), node density ( ), distance to edge node ( ). The reason behind the selection of these parameters as follow: relate to IoT node becoming a CH having enough energy to successfully execute the operation of clustering.
relates to the number of neighboring IoT nodes corresponding to the selected node as CH, if is in sparse of nature to the CH, then communication cost increases. Moreover, the worst-case cluster member's nodes exhaust all energy in the transmission of observed information because it is far from CH.
defining the CH should be at an optimal position so that it manages the communication cost both from cluster member's nodes and edge nodes.  Table 2. UT AE MD AR

Adapted Fuzzy Logic Interference System for Green Communication
In the first optimization phase, a tentative number of CHs is selected based on three parameters in an IoT network: residual energy (I re ), node density (I nd ), distance to edge node (I de ). The reason behind the selection of these parameters as follow: I re relate to IoT node becoming a CH having enough energy to successfully execute the operation of clustering. I nd relates to the number of neighboring IoT nodes corresponding to the selected node as CH, if I nd is in sparse of nature to the CH, then communication cost increases. Moreover, the worst-case cluster member's nodes exhaust all energy in the transmission of observed information because it is far from CH. I de defining the CH should be at an optimal position so that it manages the communication cost both from cluster member's nodes and edge nodes.
The linguistic variable for mamdani type FLIS as input:  Table 2. generate single output. This can be done by using a fuzzy minimum AND operator f the selection of three input MF parameters. (iii) Aggregation-as there is multiple ou value generated corresponding to 27 rules, to aggregate as one single output, maxim union operator OR is used. (iv) Defuzzification-the center of area is used for defuz cation of aggregated value into a crisp output. After the selection of CHs process, the maining nodes join the nearest CHs through the join message.

Adapted Chicken Swarm Genetic Optimization for Green Communication
In the second optimization phase, the set of tentative CHs resulting from FLIS o mization serves as the initial population of the CSGO to generate a better set of CH c pared to the first FLIS optimization phase. In the proposed CSGO algorithm, binary resentation is used to represent the IoT nodes as CH (1) or a normal IoT node (0). binary index ( ) value of an IoT node can be calculated using the sigmoid functio follows: represent the sigmoid function. The CSGO inclu adapting GA crossover and mutation processes into traditional chicken swarm optim tion to enhance the diversity in the population. In the presented algorithm, each clust represented as pack classified into Rooster, hens and chicks. Chickens have the best fitn value represented as Rooster (CHs) and chicks are the chickens with the worst fitn value. Most chickens are hens labeled as normal IoT nodes. The hens and chicks f mother-child relationships arbitrarily. Moreover, the mother-child relationship and d inance relation are unaltered in a pack and updated every distinct (θ) swarm upda The FLIS evaluates the output probability in four steps as follows: (i) Fuzzificationthis step creates MFs of the crisp input variables according to intersection point. (ii) Fuzzy rule-base-all 27 If-Then rules executed parallel on the given three input variables to generate single output. This can be done by using a fuzzy minimum AND operator from the selection of three input MF parameters. (iii) Aggregation-as there is multiple output value generated corresponding to 27 rules, to aggregate as one single output, maximum union operator OR is used. (iv) Defuzzification-the center of area is used for defuzzification of aggregated value into a crisp output. After the selection of CHs process, the remaining nodes join the nearest CHs through the join message.

Adapted Chicken Swarm Genetic Optimization for Green Communication
In the second optimization phase, the set of tentative CHs resulting from FLIS optimization serves as the initial population of the CSGO to generate a better set of CH compared to the first FLIS optimization phase. In the proposed CSGO algorithm, binary representation is used to represent the IoT nodes as CH (1) or a normal IoT node (0). The binary index (b I N ) value of an IoT node can be calculated using the sigmoid function as follows: where, sigm f (I ch ) = 1/1 + e −I ch represent the sigmoid function. The CSGO includes adapting GA crossover and mutation processes into traditional chicken swarm optimization to enhance the diversity in the population. In the presented algorithm, each cluster is represented as pack classified into Rooster, hens and chicks. Chickens have the best fitness value represented as Rooster (CHs) and chicks are the chickens with the worst fitness value.
Most chickens are hens labeled as normal IoT nodes. The hens and chicks form mother- child relationships arbitrarily. Moreover, the mother-child relationship and dominance relation are unaltered in a pack and updated every distinct (θ) swarm updating frequency as time steps. The position of the rooster is updated using rand 0, σ 2 Gaussian distribution with mean zero and standard deviation σ 2 as follow: where, Y t+1 m,D represent the position of mth rooster in the D-dimension space, {(F r ) define the fitness function of the randomly selected rth rooster and £ is the smallest constant to avoid zero-error division. The position of the mth hens' is updated as follows: where, ρ 1 , ρ 2 ∈ [1, 2 . . . .N] are the index of randomly chosen rooster and chickens (hens' or chicks), respectively, such that ρ 1 = ρ 2 . The rand( ) function generates a uniform random number between 0 and 1. Similarly, the position of mth chicks follows their mother position (FL) as follows: where, Y t M,D stands for the position of mth chick's mother, such that M ∈ [1, n]. FL( ) function allows chick's to choose any random value between zero and two. The proposed HIOA algorithm (refer to Algorithm 1) for the generation of the best set of CH works in two phases shown in Figure 3. The first phase generates the tentative list of CHs using FLIS. The second phase (CSGO) generates the optimum set of (rooster) CHs and genetic operators such as crossover and mutation are applied over hens' and chicks to enhance their fitness value. In addition, CSGO includes three steps: (i) initialization step-set the K number of CHs obtained from FLIS optimization, number of rooster (N r ), hens (N h ) and chicks (N c ), population size (pop size ), maximum number of swarm updating frequency (θ), single point crossover (ϑ) and mutation rate (τ) with maximum iteration. (ii) Selection phaseit updates the position of rooster, hens' and chicks using Equations (7)- (12). Moreover, crossover and mutation operators were applied over hens' and chicks to enhance their fitness value. (iii) Output step-individuals having the best fitness value generated as the optimum set of CHs.
Read network configuration 3.
Input: I re , I nd , I de 5.
Execute the FLIS engine based on rules define in Table 2 6.
Return the chance of an IoT node selected as CH 7.
Output: Tentative set W number of CH 8.
Second phase: CSGO algorithm 9. Initialization: Initialize the population matrix U by random values from 0 and 1. c.
Include the output of FLIS (set of k number of CHs) as one feasible solution.
10. Selection: 11. Reconstruction of infeasible solution takes places those have CHs less than k. 12. Calculate the fitness value of each row of U. 13. Optimum set (O CH ) = row of U having best fitness value. 14. For t = 1 to t max do 15.
Sort the individuals into ascending order according to their fitness value.

b.
Divide into three category rooster (N r ), hens'(N h ) and chicks (N c ). c.
Determine the relationship between mother-child in a pack.

End If
17. for each m th individuals in each Y row of U do // pop size 18.
for D = N r + 1 to pop size do 25.
generate two child offspring y o1 and y o1 from two parent chromosome Y D and Y D+1 26.
end for // end of for loop in line no. 22 28. // Mutation 29.

Operational Procedure for HIOA
An operational period for the presented HIOA algorithm operates in a setup and steady-state period that is repeated in each round. Thus, each IoT node has the chance to play the role of CHs in blanching the energy consumption in the IoT network. Firstly, in the setup period, CH election and binding of IoT nodes subject to cluster formation takes place. To get knowledge about IoT nodes, edge nodes advertise beacon messages over the network. In turn, IoT nodes respond with their ID, co-ordinate and residual energy. Using the RSSI model, IoT nodes calculate their distance from the edge node by finding their coordinate ( , ) as follow:

Operational Procedure for HIOA
An operational period for the presented HIOA algorithm operates in a setup and steady-state period that is repeated in each round. Thus, each IoT node has the chance to play the role of CHs in blanching the energy consumption in the IoT network. Firstly, in the setup period, CH election and binding of IoT nodes subject to cluster formation takes place. To get knowledge about IoT nodes, edge nodes advertise beacon messages over the network. In turn, IoT nodes respond with their ID, co-ordinate and residual energy. Using the RSSI model, IoT nodes calculate their distance from the edge node by finding their co-ordinate (X l , Y l ) as follow: where, ϕ represents the coefficient of signal propagation parameter alias exponent, z ie is the usual distance between edge node to an IoT node and A denotes the obtained signal strength in one meter distance without obstacle. Further, the edge node uses the presented HIOA for the election of K number of CHs. Furthermore, edge nodes send messages to each IoT node in the K pack to inform nodes as CHs. Each IoT node in the pack advertises its role as cluster member CM or CHs containing its ID. Those remaining nodes that do not belong to any pack choose the nearby CH based on minimum communication cost through the join message. Second, in the steady state period, the major goal is to avoid data collisions occurring during the gathering of data by parent CH from its child CM nodes. Then, CHs schedule the data gathering using the TDMA technique. Now, CM sends their observed data to their respective CH that include ID and residual energy. This local information is useful in deciding on the role of CH for the next round in a pack. CH applies a data aggregation algorithm to remove redundancy in the gathered data. Finally, CH sends a packet of fixed size to the edge node with the local information of CM's.

Results and Discussion
The performance of the HIOA algorithm is compared with a state-of-the-art algorithm using MATLAB 2019A. This section is divided into two parts: (i) CSGO parameter analysis and (ii) network parameter analysis. In addition, 200 IoT nodes are randomly distributed over the IoT network size 150 × 150 m 2 with edge node set at the corner position of the network. The initial power of the nodes is 1 J, similar to the first order radio model used in paper [20]. CSO is initializing with a random solution in the search space and applied to minimize the fitness function that yields a single solution with all features selected as the initial solution. For CSO, the maximum generation is set to be 60, the number of roosters is between 0.1 to 0.4 percentages randomly, and the number of hens is between 0.2 to 0.8 randomly. The list of other parameters and their values is listed in Table 3, which is similar to paper [29,31].

CSGO Parameter Analysis
In this subsection, the consequences of four different parameters on the fitness performance of the presented algorithm HIOA are analyzed. These parameters are (i) population size (pop size ) (ii) number of rooster (N r ) (iii) number of hens' (N h ) and (iv) swarm updating frequency θ. In addition, the number of chicks can be evaluated as N c = N − N r − N h .
From the above analysis of the result from Figure 4a-d, we observed that the presented algorithm achieves a better fitness value of about 0.625 within 60 iterations and after that more iteration, the variation in fitness value is negligible. This can be attributed to the reason the presented algorithm evaluates the fitness value on the basis of three optimization functions: inter-cluster, intra-cluster distance and energy consumption. This reveals that the presented algorithm is able to balance the energy consumption and average distance properly. Thus, the best fitness value of the proposed algorithm HIOA is obtained through setting up the parameters pop size = 60, N r = 0.3, N h = 0.6 and θ = 10. In addition, the number of chicks can be evaluated as N c = 1 − 0.3 − 0.6 = 0.1. In the next upcoming simulation, the above parameters are set with their values. From the above analysis of the result from Figure 4a-d, we observed that the presented algorithm achieves a better fitness value of about 0.625 within 60 iterations and after that more iteration, the variation in fitness value is negligible. This can be attributed to the reason the presented algorithm evaluates the fitness value on the basis of three optimization functions: inter-cluster, intra-cluster distance and energy consumption. This reveals that the presented algorithm is able to balance the energy consumption and average distance properly. Thus, the best fitness value of the proposed algorithm HIOA is obtained through setting up the parameters = 60, = 0.3, = 0.6 and = 10. In addition, the number of chicks can be evaluated as = 1 − 0.3 − 0.6 = 0.1. In the next upcoming simulation, the above parameters are set with their values.

Network Parameter Analysis
In this section, a comparison of the HIOA with LEACH-FL [27], CRGA [29] and EC-PSO [28] algorithms has been analyzed to show the effectiveness in terms of network lifetime, average energy consumed per node at each round and the total average energy consumed per round.

Comparison of Active Nodes over Rounds
In Figure 5, the number of active IoT nodes is plotted per round to show the effectiveness of the proposed algorithms compared to state-of-the-art algorithms. It is evident from the result for the starting phase of all the algorithms, up to 300 rounds around 98% IoT nodes are alive/active except in LEACH-FL only 87%. Further, on enhancement in rounds, the proposed algorithm shows the best performance compared to EC-PSO, CRGA and LEACH-FL. This is due to the fact that each chicken in the proposed algorithm acts as an agent, which updates its position in the network areas to become CH or not. In addition, this agent (chicken) has information about the whole group (cluster) and fairly balances the energy consumption load using GA crossover and mutation operator. Further, the EC-PSO algorithm does not have the ability to multi-swarm optimization and lacks adaptation to changes in the network. Whereas, CRGA do not has agent so that keep

Network Parameter Analysis
In this section, a comparison of the HIOA with LEACH-FL [27], CRGA [29] and EC-PSO [28] algorithms has been analyzed to show the effectiveness in terms of network lifetime, average energy consumed per node at each round and the total average energy consumed per round.

Comparison of Active Nodes over Rounds
In Figure 5, the number of active IoT nodes is plotted per round to show the effectiveness of the proposed algorithms compared to state-of-the-art algorithms. It is evident from the result for the starting phase of all the algorithms, up to 300 rounds around 98% IoT nodes are alive/active except in LEACH-FL only 87%. Further, on enhancement in rounds, the proposed algorithm shows the best performance compared to EC-PSO, CRGA and LEACH-FL. This is due to the fact that each chicken in the proposed algorithm acts as an agent, which updates its position in the network areas to become CH or not. In addition, this agent (chicken) has information about the whole group (cluster) and fairly balances the energy consumption load using GA crossover and mutation operator. Further, the EC-PSO algorithm does not have the ability to multi-swarm optimization and lacks adaptation to changes in the network. Whereas, CRGA do not has agent so that keep information's of other chromosomes in a population. It is also worth noting that the number of active nodes in LEACH-FL is only 20 after crossing 800 rounds and further increments in the number of rounds in turn all the nodes die. This result confirms that LEACH-FL is not able to achieve fairness in network load balance and residual energy of IoT nodes. Thus, overall, the proposed algorithm HIOA outperforms EC-PSO, CRGA and LEACH-FL by 12%, 19% and 29%, respectively, in terms of the number of active nodes remaining in the 900 rounds.
information's of other chromosomes in a population. It is also worth noting that the number of active nodes in LEACH-FL is only 20 after crossing 800 rounds and further increments in the number of rounds in turn all the nodes die. This result confirms that LEACH-FL is not able to achieve fairness in network load balance and residual energy of IoT nodes. Thus, overall, the proposed algorithm HIOA outperforms EC-PSO, CRGA and LEACH-FL by 12%, 19% and 29%, respectively, in terms of the number of active nodes remaining in the 900 rounds.

Comparison of Network Lifetime over Rounds
A comparison of the network lifetime of the proposed algorithm and state-of-the-art algorithms is shown in Figure 6. The network lifetime is the time at which the first node dies (FND) or half of the nodes die (HND) and the last node dies (LND). It can be clearly observed from the result the proposed algorithm HIOA achieves better lifetime in terms of FND is 11%, 18% and 48% from EC-PSO, CRGA and LEACH-FL, respectively. Further, HND is 27%, 35% and 58% compared with EC-PSO, CRGA and LEACH-FL, respectively. Furthermore, in terms of LND is 35%, 46% and 81% compared to EC-PSO, CRGA and LEACH-FL, respectively. This can be attributed to the reason that the proposed algorithm uses both fuzzy logic optimization with a chicken swarm genetic algorithm that helps in building a more feasible solution rather than only using particle swarm optimization in EC-PSO or a genetic algorithm in CRGCA. This is also observed from the result that LEACH-FL shows the worst performance in terms of lifetime, where FND, HND and LND occur at 100, 336 and 982 rounds. It is due to the reason LEACH-FL selects the CH based only on residual energy and node distance, where node density is left out. There is no optimization, such as PSO, or GA is used to enhance performance.

Comparison of Network Lifetime over Rounds
A comparison of the network lifetime of the proposed algorithm and state-of-the-art algorithms is shown in Figure 6. The network lifetime is the time at which the first node dies (FND) or half of the nodes die (HND) and the last node dies (LND). It can be clearly observed from the result the proposed algorithm HIOA achieves better lifetime in terms of FND is 11%, 18% and 48% from EC-PSO, CRGA and LEACH-FL, respectively. Further, HND is 27%, 35% and 58% compared with EC-PSO, CRGA and LEACH-FL, respectively. Furthermore, in terms of LND is 35%, 46% and 81% compared to EC-PSO, CRGA and LEACH-FL, respectively. This can be attributed to the reason that the proposed algorithm uses both fuzzy logic optimization with a chicken swarm genetic algorithm that helps in building a more feasible solution rather than only using particle swarm optimization in EC-PSO or a genetic algorithm in CRGCA. This is also observed from the result that LEACH-FL shows the worst performance in terms of lifetime, where FND, HND and LND occur at 100, 336 and 982 rounds. It is due to the reason LEACH-FL selects the CH based only on residual energy and node distance, where node density is left out. There is no optimization, such as PSO, or GA is used to enhance performance.
information's of other chromosomes in a population. It is also worth noting that the number of active nodes in LEACH-FL is only 20 after crossing 800 rounds and further increments in the number of rounds in turn all the nodes die. This result confirms that LEACH-FL is not able to achieve fairness in network load balance and residual energy of IoT nodes. Thus, overall, the proposed algorithm HIOA outperforms EC-PSO, CRGA and LEACH-FL by 12%, 19% and 29%, respectively, in terms of the number of active nodes remaining in the 900 rounds.

Comparison of Network Lifetime over Rounds
A comparison of the network lifetime of the proposed algorithm and state-of-the-art algorithms is shown in Figure 6. The network lifetime is the time at which the first node dies (FND) or half of the nodes die (HND) and the last node dies (LND). It can be clearly observed from the result the proposed algorithm HIOA achieves better lifetime in terms of FND is 11%, 18% and 48% from EC-PSO, CRGA and LEACH-FL, respectively. Further, HND is 27%, 35% and 58% compared with EC-PSO, CRGA and LEACH-FL, respectively. Furthermore, in terms of LND is 35%, 46% and 81% compared to EC-PSO, CRGA and LEACH-FL, respectively. This can be attributed to the reason that the proposed algorithm uses both fuzzy logic optimization with a chicken swarm genetic algorithm that helps in building a more feasible solution rather than only using particle swarm optimization in EC-PSO or a genetic algorithm in CRGCA. This is also observed from the result that LEACH-FL shows the worst performance in terms of lifetime, where FND, HND and LND occur at 100, 336 and 982 rounds. It is due to the reason LEACH-FL selects the CH based only on residual energy and node distance, where node density is left out. There is no optimization, such as PSO, or GA is used to enhance performance.

Comparison of Average Energy Consumed over Rounds
A comparison of average consumed energy over rounds of the proposed algorithm HIOA and state-of-the-art algorithms is shown in Figure 7. Initially, all the state-of-the-art algorithms except LEACH-FL [27], up to 300 rounds, consume almost similar amounts of energy. Thereafter, on the enhancement of the number of rounds, the proposed algorithm HIOA consumes much less energy than other state-of-the-art algorithms and goes up to above 1200 rounds with a consumption of 0.75 Joule. In addition, it can be also observed that CRGA [29] and EC-PSO [28] consume almost 0.95 Joule and 0.85 Joule of energy within 972 rounds and 1030 rounds, respectively. This is due to the fact presented HIOA algorithm inherit CSGO that designated to inmate multi-swarm optimization group with crossover and mutation technique of genetic algorithm, which enhances the efficient use of energy in the IoT network. EC-PSO is a single group swarm optimization technique that is lacking in the generation of feasible solutions of CHs throughout the IoT network. In addition, the fitness function of CRGA only includes residual energy parameter and left the inter-cluster distance and intra-cluster distance parameter, so that selection of CHs is not optimal. Thus, overall, the proposed algorithm HIOA outperforms EC-PSO, CRGA and LEACH-FL by 17%, 26% and 38%, respectively.

Comparison of Average Energy Consumed over Rounds
A comparison of average consumed energy over rounds of the proposed algorithm HIOA and state-of-the-art algorithms is shown in Figure 7. Initially, all the state-of-the-art algorithms except LEACH-FL [27], up to 300 rounds, consume almost similar amounts of energy. Thereafter, on the enhancement of the number of rounds, the proposed algorithm HIOA consumes much less energy than other state-of-the-art algorithms and goes up to above 1200 rounds with a consumption of 0.75 Joule. In addition, it can be also observed that CRGA [29] and EC-PSO [28] consume almost 0.95 Joule and 0.85 Joule of energy within 972 rounds and 1030 rounds, respectively. This is due to the fact presented HIOA algorithm inherit CSGO that designated to inmate multi-swarm optimization group with crossover and mutation technique of genetic algorithm, which enhances the efficient use of energy in the IoT network. EC-PSO is a single group swarm optimization technique that is lacking in the generation of feasible solutions of CHs throughout the IoT network. In addition, the fitness function of CRGA only includes residual energy parameter and left the inter-cluster distance and intra-cluster distance parameter, so that selection of CHs is not optimal. Thus, overall, the proposed algorithm HIOA outperforms EC-PSO, CRGA and LEACH-FL by 17%, 26% and 38%, respectively.

Comparison of Average Residual Energy over Rounds
A comparison of the average residual energy per IoT node with respect to the round between the proposed algorithms and state-of-the-art algorithms is shown in Figure 8. The average residual energy is calculated as dividing the overall residual energy per round by the total number of IoT nodes. It is obvious that the residual energy of IoT nodes that uses the presented HIOA has more residual energy (0.12 Joule) after completion of 1200 rounds, whereas CRGA [29] and EC-PSO [28] algorithms consume more energy in data communication and IoT nodes have only 0.05 Joule and 0.02 Joule of energy art 1000 rounds. This is due to the fact that the proposed algorithm HIOA optimally selects the CHs at each round and balances the energy consumption load among all the nodes within a cluster. This can be attributed to the proper division of IoT nodes in rooster, hens and chicks, where the mother-child (hens & chicks) relationship is also balanced. In addition, crossover and mutation operators also help in enhancing the selection of optimum set of CHs. It is also worth noting that initially the LEACH-FL [27] algorithm consumes less energy and IoT nodes have a handsome amount of residual energy left out. However, in the increment of rounds, the residual energy of IoT nodes declined sharply and almost zero within 892 rounds. This is because LEACH-FL ignores the inter-cluster and intracluster communication costs in the selection of CHs that consume extra energy. Thus, overall, the proposed algorithm HIOA has more residual energy compared to EC-PSO, CRGA and LEACH-FL by 11%, 16% and 21%, respectively.

Comparison of Average Residual Energy over Rounds
A comparison of the average residual energy per IoT node with respect to the round between the proposed algorithms and state-of-the-art algorithms is shown in Figure 8. The average residual energy is calculated as dividing the overall residual energy per round by the total number of IoT nodes. It is obvious that the residual energy of IoT nodes that uses the presented HIOA has more residual energy (0.12 Joule) after completion of 1200 rounds, whereas CRGA [29] and EC-PSO [28] algorithms consume more energy in data communication and IoT nodes have only 0.05 Joule and 0.02 Joule of energy art 1000 rounds. This is due to the fact that the proposed algorithm HIOA optimally selects the CHs at each round and balances the energy consumption load among all the nodes within a cluster. This can be attributed to the proper division of IoT nodes in rooster, hens and chicks, where the mother-child (hens & chicks) relationship is also balanced. In addition, crossover and mutation operators also help in enhancing the selection of optimum set of CHs. It is also worth noting that initially the LEACH-FL [27] algorithm consumes less energy and IoT nodes have a handsome amount of residual energy left out. However, in the increment of rounds, the residual energy of IoT nodes declined sharply and almost zero within 892 rounds. This is because LEACH-FL ignores the inter-cluster and intra-cluster communication costs in the selection of CHs that consume extra energy. Thus, overall, the proposed algorithm HIOA has more residual energy compared to EC-PSO, CRGA and LEACH-FL by 11%, 16% and 21%, respectively.

Comparison of Standard Deviation over Rounds
A comparison of the standard deviation of residual energy and energy load balance by the CHs between the proposed algorithm and state-of-the-art algorithms is shown in Figure 9a and Figure 9b, respectively. The standard deviation of residual energy measures the variability and consistency of the residual energy of the IoT nodes in the population. This is evident from the result that the standard deviation of residual energy and energy load of the selected CHs for the proposed HIOA algorithms is more balanced compared to EC-PSO [28], CRGA [29] and LEACH-FL [27]. This can be attributed to the reason that it's minimizing the energy consumption during the selection of the fitness function of CHs by considering inter-cluster and intra-cluster distance. In addition, GA operators (crossover and mutation) help in reducing the convergence time in selecting the global optimum set of CHs, which increases population diversity and not to select local optimum CH sets. EC-PSO forms the clusters according to swarm particle optimization, and normal nodes join the cluster according to the distance from the edge node and left behind the number of neighboring IoT nodes. This can be attributed to unbalanced energy consumption in the cluster. It can also be noticed that CRGA and LEACH-FL show variations in the standard deviation of the IoT node's residual energy and load balancing by CHs. It is due to the fact both CRGA and LEACH-FL algorithms choose the CHs randomly, in turn, unevenly distributing the load on the CHs and reducing residual energy. The worst performance is shown by LEACH-FL because probabilistic selection generates some isolated CHs that consume much more energy in the transmission of data to edge nodes. Thus, the overall

Comparison of Standard Deviation over Rounds
A comparison of the standard deviation of residual energy and energy load balance by the CHs between the proposed algorithm and state-of-the-art algorithms is shown in Figure 9a and b, respectively. The standard deviation of residual energy measures the variability and consistency of the residual energy of the IoT nodes in the population. This is evident from the result that the standard deviation of residual energy and energy load of the selected CHs for the proposed HIOA algorithms is more balanced compared to EC-PSO [28], CRGA [29] and LEACH-FL [27]. This can be attributed to the reason that it's minimizing the energy consumption during the selection of the fitness function of CHs by considering inter-cluster and intra-cluster distance. In addition, GA operators (crossover and mutation) help in reducing the convergence time in selecting the global optimum set of CHs, which increases population diversity and not to select local optimum CH sets. EC-PSO forms the clusters according to swarm particle optimization, and normal nodes join the cluster according to the distance from the edge node and left behind the number of neighboring IoT nodes. This can be attributed to unbalanced energy consumption in the cluster.

Comparison of Standard Deviation over Rounds
A comparison of the standard deviation of residual energy and energy load balance by the CHs between the proposed algorithm and state-of-the-art algorithms is shown in Figure 9a and Figure 9b, respectively. The standard deviation of residual energy measures the variability and consistency of the residual energy of the IoT nodes in the population. This is evident from the result that the standard deviation of residual energy and energy load of the selected CHs for the proposed HIOA algorithms is more balanced compared to EC-PSO [28], CRGA [29] and LEACH-FL [27]. This can be attributed to the reason that it's minimizing the energy consumption during the selection of the fitness function of CHs by considering inter-cluster and intra-cluster distance. In addition, GA operators (crossover and mutation) help in reducing the convergence time in selecting the global optimum set of CHs, which increases population diversity and not to select local optimum CH sets. EC-PSO forms the clusters according to swarm particle optimization, and normal nodes join the cluster according to the distance from the edge node and left behind the number of neighboring IoT nodes. This can be attributed to unbalanced energy consumption in the cluster. It can also be noticed that CRGA and LEACH-FL show variations in the standard deviation of the IoT node's residual energy and load balancing by CHs. It is due to the fact both CRGA and LEACH-FL algorithms choose the CHs randomly, in turn, unevenly distributing the load on the CHs and reducing residual energy. The worst performance is shown by LEACH-FL because probabilistic selection generates some isolated CHs that consume much more energy in the transmission of data to edge nodes. Thus, the overall It can also be noticed that CRGA and LEACH-FL show variations in the standard deviation of the IoT node's residual energy and load balancing by CHs. It is due to the fact both CRGA and LEACH-FL algorithms choose the CHs randomly, in turn, unevenly distributing the load on the CHs and reducing residual energy. The worst performance is shown by LEACH-FL because probabilistic selection generates some isolated CHs that consume much more energy in the transmission of data to edge nodes. Thus, the overall result proved that the standard deviation of residual energy of the proposed algorithm HIOA was lower than the EC-PSO, CRGA and LEACH-FL by 18%, 26% and 42%, respectively. The load balancing of the proposed algorithm is also 23%, 32% and 48% lower than state-of-the-art algorithms.

Conclusions and Future Scope
In this paper, we proposed a new hybrid algorithm for clustering in the IoT network based on FL and CSO employing a GA to minimize energy dissipation. To this end, tentative CHs are selected using FL by considering essential parameters. Further, enhanced CSO with GA utilizes the concept of hierarchal order (rooster →hen →chicks) to divide the population and their fitness values are evaluated in order to find the best of nodes as CHs using crossover and mutation operation of GA. Simulations conclude that our proposed algorithm HIOA is robust to different CSO parameters and finds a near-optimal solution with low time complexity and higher convergence speed over state-of-the-art algorithms. The particle implementation of the simulate model can be used for load balancing or self-organizing structure in IoT nodes (electric vehicles) for the energy buying market. The presented HIOA model effectively captures the rapid and frequently changing traffic patterns due to the inherent feature of GA and the dynamic classification property of CSO with a lower rate of latency for convergence. As the proposed model has no prediction ability for a sustainable network where electric vehicles are used as nodes, to add these features in the future, we included a neural network or deep learning approach for the formation of clusters based on past experience [34]. The nearby models include training and testing and deployment as well in different network scenarios.