UWSNs were divided into sensor nodes (SNs) and data collection stations (collectively referred to as sinks) (Awan et al., 2019; Bharany et al., 2021; Khan et al., 2021). Sensor nodes are small devices that track and demonstrate physical characteristics and transmit the information to a base station using wireless communication protocols. Radiofrequency waves are not suggested in underwater sensor networks because their signal attenuation is excessive and limited in distance range (Awan et al., 2019; Khan et al., 2021; Nguyen et al., 2021). Due to the limited capacity and frequency of the UW channel, its bandwidth is significantly smaller than that of a conventional radio channel. Due to the high absorption, attenuation, and propagation, UWSNs have a restricted bandwidth, which piqued researchers’ curiosity. Another important consideration is the failure of SNs, which might be due to environmental conditions or some other. Energy conservation is another major limitation of UWSNs that significantly impacts network longevity. There is no way to replace or recharge network nodes in the harsh environment of the UWSN, which means the network will run out of nodes (Bharany et al., 2021) with time. Tsunami predictions, offshore exploration, tactical surveillance, oil and gas spill monitoring, aided navigation, pollution monitoring, and several commercial uses are all possible with UWSNs, which are further visualized to enable a wide range of applications (Nguyen et al., 2021). Underwater devices need to communicate with each other optimally for these applications to work efficiently (Bharany et al., 2021). Differentiating from the terrestrial environment, the marine environment is far more complicated and variable, making ocean monitoring more challenging. A UWSN is an ad-hoc network designed for underwater communication and data transmission in which several surface stations are required for better communication (Awan et al., 2019; Bharany et al., 2021; Khan et al., 2021; Nguyen et al., 2021) so that Sensor nodes can easily transmit and receive data between a source and a destination.

Sensor nodes are responsible for three primary functions: data monitoring, processing, and reception of data. Each node is designed to sense and monitor a particular physical or chemical characteristic underwater and transmit the information gathered to a central point (Heinzelman et al., 2002). The following are some of the most significant issues related to undersea applications:

Electronic Magnetic waves at high frequencies suffer from considerable attenuation due to the water’s saltiness. They cannot be used underwater because of their high frequency (Reddy and Khare, 2017). Several obstacles still make underwater communication more difficult, even after considering the mentioned challenges (Wan et al., 2018). One such issue is a need for more available bandwidth combined with significant energy usage. It has an immediate impact on the transmission rate of a network. UWSNs use energy quickly due to the current in the water and the many sources of noise present. Because of these issues, clustering in UWSNs has become problematic and expensive (Zhang et al., 2017). Developing an effective clustering procedure to handle the issues outlined above is necessary. The unique characteristics of UWSNs have attracted the attention of researchers interested in exploring their designs and networking protocols. Grounded on this interest, we developed an energy-efficient technique for clustering UWSNs that outperformed LEACH and other bio-inspired clustering algorithms (Heinzelman et al., 2002; Singh and Lobiyal, 2012; Rao et al., 2017; Alhazmi et al., 2018; Wan et al., 2018). Among the most well-known and energy-efficient clustering algorithms, those employed in terrestrial wireless sensor networks cannot deploy in the UWSN network. We also try to figure out why the WSN clustering protocols do not work well in the UWSN (Heinzelman et al., 2002).

According to the proposed technique, UWSN nodes will be separated into clusters using a dynamic cluster head selection process, as seen in Figure 2 . In sensor network design, clustering organizes sensor nodes into logical groups called a cluster. Several nodes assemble information and transfer it to the central station as the Base station. Only one node in the relay is designated as the cluster head, while the leftover nodes are referred to as member nodes (Nguyen et al., 2021). The cluster head’s responsibility is to ensure that all inter- and intra-cluster communication is appropriately coordinated. When choosing the cluster head for each node, the amount of energy left and other parameters of the nodes will be considered. Additionally, the distance between CH and BS may be considered in this decision. The cluster size and the cluster heads’ election are already important parameters for decision-making (Kazmi et al., 2019).

Additionally, the correct number of clusters in UWSNs impacts the universal energy performance of the network. It will be necessary to select a new CH when the energy of the cluster head is less than a particular threshold value. Optimizing the glowworm swarm optimization (GSO) algorithm (Javadpour et al., 2018; Reddy and Khare, 2017; Rajakumar et al., 2017; Potthuri et al., 2018) is the basis for developing the proposed protocol. An evolutionary-based optimizer ensures enough clusters to keep the network running for a long time, resulting in fewer hops and lower energy usage.

The following are the most significant contributions made by this research:

Underwater sensor networks have emerged as a new area of interest for academics looking for new and novel approaches. Several researchers have discovered their distinguishing characteristics for optimal network communication and have handled the setup challenges that arise (Javadpour et al., 2018). Compared to wireless sensor networks, the underwater medium presents an exceptionally high degree of unpredictability and difficulty. The most significant distinctions that make UWSN different from WSN are as follows: (1) amount of energy consumption; (2) propagation delay; (3) a low bandwidth; (4) dynamic topology operation; (5) propagation speed; (6) efficiency; (7) data transmission rate; and (8) noise interference. Several academics have developed a new energy-efficient clustering methodology for UWSNs (Nguyen et al., 2021). Because of the specific constraints of the environment of the underwater scene, the clustering technique developed for terrestrial networks cannot be easily used for UWSNs. Conserving energy is a critical goal for UWSNs. The clustering method is a well-known technology that has been successfully applied to cut down the energy usage of UWSNs (Bharany et al., 2021). A considerable amount of research has been done on clustering and the election of cluster heads to lower the energy consumption of the UWSN. However, there needs to be more research carried out on UWSN. This part addresses some of the research findings in even more detail.

The research in (Heinzelman et al., 2000) proposed a Low-Energy Adaptive Clustering Hierarchy as a low-energy replacement for traditional clustering techniques. Selecting cluster leaders randomly and cycling between them is a common tactic in hierarchical networks to ensure fair energy distribution. LEACH considers the nodes’ available data, including their residual energy, communication energy usage, and the number of neighboring nodes, to select the most energy-efficient CH nodes. The LEACH-C algorithm is a significant improvement over its forerunner. Sending information about the sensor nodes’ locations and remaining battery life to the base station can increase the system’s overall efficiency. Centralized deployment of LEACH-C lowers the network’s total energy consumption and each node’s energy consumption. However, LEACH-C also has some problems. If we take the LEACH-C network as an example, every node has an equal opportunity to become a CH. If the network runs out of power, the least powerful nodes will have to act as CHs, reducing the network’s overall efficiency and placing an undue burden on the system. To create an energy-conscious CH selection approach, Singh et al. (Singh and Lobiyal, 2012) employed PSO for clustering the nodes. The fitness function is derived by multiplying the fitness metric by the sum of the residual energy, the distance, and the node density. This approach requires consideration of inefficient cluster formation, which drains the network’s resources. In order to solve the issue brought up in (Heinzelman et al., 2002; Singh and Lobiyal, 2012), Rao et al. (2017) presented a PSO-based energy-efficient cluster head selection (PSOECHS) that includes journey time to the base station (BS) in its fitness function.

PSOECHS also affects the CH selection mechanism by changing the total number of nodes in each cluster, which extends the network’s lifespan. Contrarily, computing and selecting CHs at each node consumes much more power than that. Known as UMOD-LEACH (Alhazmi et al., 2018), it is a highly-improved variant of the Leach protocol. When applied to water, this technique outperforms LEACH by 30%. It uses localization and time-division multiple-access (TDMA). According to (Wan et al., 2018), the Adaptive Clustering Underwater Network (ACUN) for underwater networks was a successor to the aforementioned adaptive clustering routing system. The amount of remaining energy at the CH node and the distance between the CH node and the sending node are the two primary factors in multipath routing.

Zhang et al. (2017) developed a clustering scheme for a medium-sized UWSN using discrete PSO and a genetic algorithm (GA) to increase the network’s durability. However, the methodology is not advised because of its low stability, and the clustering model needs to be more precise to use UWSN. Pengwei Li et al. (2017) created an improved particle swarm optimization technique for clustering UWSNs that considers individual nodes’ transmission power needs. However, fixing the issues from the previous study and having the network last even longer is challenging because particle coding is much more complex for sensor nodes. The Energy Center Searching model employing PSO (EC-PSO) was proposed in the literature (Wang et al., 2019) to prevent these energy gaps and to search for energy centers for CHs selection. Extra measures were taken to ensure that no two CHs would ever be close to one another. A safeguard with a threshold value was implemented to prevent low-energy nodes from delivering data. However, because the distance between nodes, CHs, and the sink node is not considered while making the fitness function, it affects how much energy the system uses.

The fuzzy enable clustering strategy was used to improve the routing method. In the literature (Alia, 2014), it has been suggested that decentralized fuzzy C-means are used as part of an energy-efficient routing protocol to reduce the total amount of energy a network uses while also making it last longer. CHs are chosen iterative through each cluster one at a time. Bhatti et al. (2016) developed the fuzzy c-means clustering and energy-efficient CH selection mechanism. The FCM method makes clusters, and CHs are chosen based on the SNs’ location, signal-to-noise ratio, and residual energy.

On the other hand, the sensor network can only be built in a two-dimensional space in terms of its topology. Researchers who worked with wireless sensor networks came up with a clustered routing method called BECR that used energy in a balanced way (Zhao et al., 2018). The first cluster head chosen was the node closest to the group’s center. When the energy level of a cluster head fell below 20%, the FCM was used to choose a new cluster head for that cluster. A fresh round of cluster head elections was performed if the energy level of each cluster head declined by 10%; otherwise, no elections for cluster heads were performed. The randomness feature makes it difficult to avoid falling into the trap of local optimality while employing this technique. So far, a variety of researchers have used the hybrid strategy, which has the advantage of resolving the weaknesses of one methodology while simultaneously incorporating the benefits of another. To generate clusters, the literature (Sharma et al., 2019) utilized FCM and selected the best node to act as a CH. It used a fuzzy inference technique to determine how well each node performed. This technique was used in conjunction with an evolutionary method to choose the CH to make better decisions. The literature (Shokouhifar and Jalali, 2017) reported using an artificial bee colony approach to update the fuzzy rules of LEACH-SF, which was subsequently corrected. It was made to double the network’s lifetime by following the rules, and the algorithm’s fitness function makes this happen.

We have widely deployed sophisticated algorithms to ensure efficient data routing in WSNs (such as the PSO, GWO, and MFO). However, now we are considering these techniques for UWSNs. In (Javadpour et al., 2018), fuzzy clustering was used to connect sensors in a network, and the initial value of the cluster heads was estimated using the PSO technique. It started by clustering using the fuzzy technique and then selected the best cluster head by applying the PSO algorithm to the fuzzy table created by the fuzzy approach. The algorithm had a considerable influence on reducing the amount of energy consumed. Afterward, various clustering methods were proposed by taking this as a base, including differential evolution and simulated annealing (Potthuri et al., 2018), fuzzy c-means, and the Genetic algorithm (Reddy and Khare, 2017). An optimization approach known as the Grey Wolf Optimization Method (GWO) is also used in many UWSNs to optimize clustering as an optimization methodology. GWO-LPWSN (Rajakumar et al., 2017) employs the grey wolf optimization technique to find nodes and fix placement faults. For wireless sensor networks (WSNs), the proposed method (Kazmi et al., 2019) offers a GWO-based transmission rate management mechanism. The grey wolf optimizer is used to fine-tune the support vector machine’s (SVM) transmission rate, which controls the rate data is sent between nodes. In particular, a cluster head is chosen by the sink node, and the optimum path for data transmission is determined by the nodes using GWO (Al-Aboody and Al-Raweshidy, 2016).

Similarly, optimal grouping is achieved with the help of grey wolf optimization and evolutionary algorithms (Lipare et al., 2020). The moth flame optimizer is only one of several innovative algorithms that can be used to solve optimization problems. Taking cues from moths in the wild can help find optimal solutions to NP-complete problems. Optimal clustering is the goal of a method proposed in (Mittal, 2018) that uses the MFO algorithm. Moth-flame and genetic algorithms are used for efficient and effective clustering (Sharma et al., 2020). Due to its high-quality results, it can also be implemented in UWSNs (Bharany et al., 2022). Another routing protocol designed for innovative ocean applications is SOSNET (Durrani et al., 2019). The moth flame optimizer (MFO) is employed to efficiently perform optimal clustering at a low cost (Patel et al., 2020; Bharany et al., 2022). Similarly, the technique proposed in (Kumari et al., 2019) addresses the problem of fault-tolerant routing by employing MFO, which determines the optimal data transmission path (Namasudra and Roy, 2018; Fattah et al., 2020; Bharany et al., 2022). A comparison of algorithms in Table 1 shows how they deal with different problems and have different goals.

The GSO algorithm assigns a luciferin value to each glowworm and a neighborhood range for making decisions locally. The objective function and its position establish the luciferin value of a glowworm. Compared to other glowworms, a higher luciferin value indicates that the glowworm is located in an exceedingly bright area. You can see the entire SS-GSO process in Figure 3 , which is the algorithm.

The following equation is used to keep track of the glowworm’s Luciferin value:

With respect to each individual glowworm i, Luci_i (t) symbolizes the luciferin value of that glowworm. In addition, the luciferin decay constant, announced by ρ which ranges from zero to one, the luciferin enhancement fraction, which is acted by γ, and F (p_i (t)) which represents the objective function, given for current position p_i of glowworm i.

Then, using the following rule, each glowworm i investigates its immediate surroundings in search of the neighbour with the greatest luciferin value:

Luci _z  (t)>Luci _i (t)

Ni(t) is the group of glowworms surrounded by glowworm i, while z is i’s nearest neighbor. Glowworm i’s local decision range is denoted by rangi(t), and the luciferin concentrations of glowworm z and i is represented by Luciz (t) and Luci i(t), respectively. Using the probabilities of each neighboring glowworm, as shown in Equation 3, a best-neighbor glowworm is selected.

The location of the glowworm is updated in accordance with the best position of an adjacent glowworm that has been identified and computed as follows:

where s > 0 denotes the distance travelled by a glowworm in order to reach another glowworm. The decision range randec_i (t) is computed by the following equation:

Where rads is the constant representing the radial sensor range, β represents the model constant, and n_t controls the maximum allowed number of neighbors.

In our proposed clustering strategy the selection of CH is considered based on connectivity with the base station and fitness function. Fitness function is contingent the luciferin value, total energy consumption, and residual energy value of the UWSN to ensure effective communication and data transfer (Khan et al., 2019).

The fitness functions are represented by the following equation:

where w₁, w₂, w₃ = 0.5

First of all, we calculate the residual energy of the every UWSN node by:

The residual energy is represented by ReSEi, the starting energy of the ith UWSN is represented by initialEi, and the current energy level of the ith UWSN is represented by CurrentE_i . After calculating residual energy of every node, we make an average of all nodes residual energy and then compare the residual energy of a node with average energy of all nodes. If current energy of a node is less than the average energy then this node is being skipped for the current round of CH selection.

Total amount of transmission and reception energy of a node is calculated by:

The number of transmitted bits and received bits in node i are denoted by the TranB_i and RecB_i, respectively. Engy_Tran and Engy_Rec are characterized as transmission energy and reception energy correspondingly are calculated as:

ET_radio and ER_radio are the energy needed for radio transmission by the transmitter and the receiver, respectively. EA is the energy needed to boost(amplify) a signal, and Dist [Khan et al., 2019; Bharany et al., 2021] is the distance between two nodes, m and n.

The function for luciferin value is given by:

In our proposed algorithm, Eq. 6 is used to calculate the fitness of each UWSN in the cluster formulation mechanism, as seen in Figure 4 . The weighted sum of the luciferin value and the total energy spent and residual energy from Equations. 8, 10, and 11 is used to compute the fitness. In addition to its fitness, each UWSN sends a “Hello Message” along with its fitness to all the other nodes. When the SNs get a Hello message, the received fitness is compared by the nodes with its own. When a “Hello” message is received, the UWSN creates and updates a neighbour table using the UWSN’s entries. The UWSN then sorts the neighbour table by decreasing fitness values. The closest node to the base station is the “Cluster Head”. This would be the best place for CH. However, when there is just one node in the range of the base station or more than one, the UWSN with the most significant fitness value is designated as the cluster head and sends a Cluster Formation message to all nearby nodes. By sending a cluster-joining message from all nodes with lower fitness values, they can recognize their CH. The UWSN coalition of ad-hoc nodes is created on-demand when a UWSN wants to transfer data but is out of range of the base station. In order to get the information to its final destination, the Nodes use an intermediate UWSN to make many hops (Khan et al., 2019; Bharany et al., 2021; Khan et al., 2021). The remainder of the UWSN nodes will become superficial nodes or cluster members after the election of the CH, which will be purely based on a fitness function. The Luciferin level values of each glowworm in each cluster are updated according to their position and fitness function. Members of the cluster must keep track of the CH’s movement and alter their locations accordingly. The CH keeps track of the cluster’s topology by receiving location data from all UWSNs and updating the topology database accordingly. Suppose a UWSN travels out of the cluster’s range based on its updated position (as determined by the luciferin value). In that case, it is no longer considered a part of the cluster. In the next step, the CH transmits the updated topology table to the cluster members and maintains the cluster (Bharany et al., 2021). This method also includes a technique for managing the topology of clusters, which is described in detail below. Using a topology configuration message, each CM communicates its luciferin value to the other CM UWSNs. Upon receipt of the topology configuration message, the cluster head of UWSN modifies the location of the cluster member nodes following the luciferin values stored in the cluster topology table. Once the position information for each CM of UWSN has been updated, the cluster head sends this cluster topology table to all the cluster members. This way, all the CM behave like a swarm and move to keep up with the CH.

The best path selection for information transmission in an efficient routing mechanism result in lower energy consumption in the network and enhances the cluster’s lifetime. Every UWSN node in the proposed system follows the location of CH and adjusts its position in response to that of CH. Once the sensor nodes are in clusters, each group’s head node (CH) makes a TDMA schedule and sends it to the other cluster members. In order to save energy, each node sends its data to the cluster head during the time slot defined in the TDMA table and then switches off its radio transmitter while waiting for the next transmission slot as stated in the TDMA table. In the TDMA table, the priority of allotting a slot is determined by the UAV’s neighbor range, residual energy, and location, which are used for load balancing and congestion management purposes. When allocating TDMA slots, the UWSN in the neighborhood range of the source UWSN (close to the source UAV) and having the lowest residual energy will be prioritized over the other UWSNs, as this would be necessary to retrieve the data from this node. Therefore, a node with low residual energy will be assigned to the first available TDMA slot before any other node with high residual energy. On the other hand, the cluster heads keep their radios turned on to receive data from cluster members, and they perform an XOR operation on that data to reduce the amount of duplicate data. As soon as a cluster head gets data from its members, it performs data aggregation and sends it to the surface station near the water’s edge. As the primary base station on the ground is typically located a long distance away, employing these strategies will prevent a high-energy transmission since sending it straight to the primary base station on the ground would result in high energy consumption and a high risk of data loss. This method sends information through the surface station to the primary base station to save energy and transmission costs.

In Figure 5 , we can see the number of alive nodes (NAN) analysis of the SS-GSO method compared to older methods is done. The experiment was carried out using the 2018a version of MATLAB programming, and it was completed on a 7th generation core i5 machine with DDR4 8 GB of RAM. An experimental grid size ranging from 500 meters to 2000 meters was used for this investigation. The number of nodes used ranged from 0 to 300, and each node’s transmission range was between 25 m and 200 m. Nodes are made to stay in the same place or move very slowly so they do not get in the water flow (Bharany et al., 2022). It was equated to different state-of-the-art evolutionary clustering algorithms, such as the ant colony optimizer (ACO), the grey wolf optimizer (GWO), the moth flame optimizer (MFO), and LEACH. In Table 2 , we can see all the simulation metrics. The experimental results suggest that the low-energy adaptive clustering hierarchy (LEACH) technique delivers poor outputs with the least feasible NAN. The MFO technique obtained a more significant NAN value than the LEACH procedure. Accordingly, the ACO techniques do not produce a favorable outcome either. While the GWO approach sought to create a respectable NAN compared to the other, Figure 5 revealed that the SS-GWO strategy outscored them all in terms of NAN. We can also see in the figure that the LND (the last node died) is best for our proposed algorithms. In Table 3 , we can see the results of the network node analysis of all the techniques.

It is common practice to refer to the computational complexity of a clustering algorithm as the time it takes to complete the clustering process using that approach. Typically, clustering algorithms input a set of nodes and associated fitness ratings and output a subset of CH nodes and other related nodes. The time it takes for an algorithm to go from receiving inputs to generating outputs is known as its “cluster building time.” Since the UWSN has little memory and processing power, a longer cluster formation time will harm WSN performance. Increasing the time required to create a cluster uses more energy, reducing the time SNs in a network can remain in operation. Figure 6 shows that our proposed SS-GSO performs better than LEACH, ACO, MFO, and GWO because, in contrast to ACO and GWO, which begin with several solutions and converge to the optimal through iteration, our SS-GSO always quickly converges towards the best solution. Using the SS-GSO method, the time it takes to build a cluster and the time it takes to choose a path is kept to a minimum. This makes it possible for SNs to do complex calculations with much less energy.

It is shown in Figure 7 that our suggested method, SS-GSO, compares favorably to other strategies in terms of total energy consumption (TEC). The LEACH procedure could have been more efficient than alternative techniques with a maximum TEC. It was also seen in Figure 7 that the MFO approach had a lower TEC than the LEACH method. Regarding TECs, MFO approaches were slightly better throughout several iterations than the LEACH method. It is worth noting that while the GSO performed better in all other elevation parameters, in TEC, the ACO strategy had the second-lowest TEC and surpassed the GWO strategy. Finally, our projected SS-GSO executes better than all other techniques (He et al., 2016; Bharany et al., 2022), as only ACO competes with SS-GSO.

The number of cluster head nodes substantially influences the efficiency of the protocol, as there will be an increase in node energy consumption and overall energy consumption. As a result, if the count of CHs is low, the number of SNs per CH will be high, and the time of SN data transmission will be affected (Okoth et al., 2022; Srivastava et al., 2022). The total energy consumed by each round of networks increases and is directly proportional to the number of CH. Figure 8 shows that our proposed SS-GSO clustering protocols’, outperform LEACH, ACO, GWO, and MFO protocols in headcount fluctuation frequency. In this case, we are considering only 1000 rounds for evaluation. It is simple to see that the cluster headcount in the LEACH protocol swings between 4 ≤ T ≤ 17 and 5 ≤ T ≤ 14 in GWO, 6 ≤ T ≤ 14 in ACO, and 3 ≤ T ≤ 17 in MFO, and that the cluster headcount in our suggested protocol swings between 3 ≤ k≤ 13 which is the most desirable.

This section compares the number of packets received (NOPR) analysis of the SS-GSO technique to other methods in various iterations. The experimental findings indicated that the LEACH strategy produced inferior results with a lower NOPR than the conventional approach. The MFO approach obtained a NOPR marginally more significant than the LEACH procedure. After that, ACO approach rounds resulted in a NOPR that was only marginally closer to the original last NOPR. While the GWO method was trying to get a reasonable NOPR, the SS-GSO method could get a higher NOPR than other methods (Mohamed et al., 2019; Arshad et al., 2020; Asif et al., 2020; Mehmood et al., 2020; Ahmad et al., 2022; Mazhar et al., 2022; Zeb et al., 2022).

As seen in Figure 9 and Table 4 , the SS-GSO approach is a good strategy for maximizing energy efficiency and longevity in the UWSN context. Overall, the results of the experiments showed that the LEACH method performed poorly across all comparison criteria (Mohamed and Mohamed, 2017; Mohamed and Suganthan, 2017; Ahmad et al., 2021; Mehmood et al., 2021; Saleem et al., 2021; Ahmad et al., 2022). Finally, our proposed method has performed better than all other techniques.