1. Introduction
Wireless sensor networks (WSNs) are networks composed of several nodes strategically deployed in the environment, where sensors are tasked with sensing and transmitting information to the sinks and base stations. However, these sensors have limited battery power and are very difficult to recharge following deployment [
1]. Therefore, the present study’s objective was to reduce the energy exhaustion of sensors and thereby extend the overall lifetime of the sensor network. To minimize network energy consumption, nodes are grouped into clusters [
2], with each cluster containing a designated cluster head collecting and aggregating information from its members. Subsequently, the cluster head forwards the collected information to the sink using either a single- or multi-hop approach. The selection of cluster heads is a challenging task, and various techniques have been employed to achieve an optimal selection of cluster heads [
3,
4].
Because the direct transmittance of data to the base station demands higher energy expenditure from the cluster heads, clustered sensor networks require a routing protocol that prioritizes minimizing energy consumption while selecting the optimal path from the clusters to the base station [
5]. However, the construction of a clustering-based routing algorithm with high energy efficiency and increased packet transmission rate for large-scale sensor networks is an NP-hard problem [
6]. In other words, the process of selecting optimal cluster heads and establishing efficient routing paths for large-scale sensor networks entails high time complexity. This issue can be efficiently solved by metaheuristic algorithms [
7,
8] and swarm intelligence [
9,
10].
This paper introduces an energy-saving clustering mechanism utilizing a chaotic genetic algorithm (CGA) coupled with the construction of an energy-efficient routing system using swarm intelligent grey wolf optimization (GWO). The proposed system, named ‘chaotic genetic algorithm–grey wolf optimization (CGA-GWO)’ is designed to minimize overall energy consumption by selecting energy-aware cluster heads and devising an optimal routing path to reach the base station. The key contributions and novelty of this study are summarized as follows:
Chaotic systems, including logistic and tent maps, are employed to generate the initial population and govern the crossover and mutation processes of the genetic algorithm for the selection of cluster heads. The fitness function is specifically designed to identify chromosomes with higher residual energy and designate them as cluster heads. To accelerate genetic algorithm convergence, optimal cluster heads are selected by the elitist selection method in place of the roulette selection method.
To determine an energy-efficient routing path to the base station, the GWO method was selected for its ability to yield optimal solutions across multiple iterations. The efficacy of these solutions was evaluated using a fitness function, where a higher fitness value signifies a reduction in overall distance, fewer hops, and minimized energy consumption along the routing path.
The remainder of this paper is organized as follows:
Section 2 presents an investigation of pertinent mechanisms concerning energy-efficient cluster-based routing, along with essential background information on CGA and GWO;
Section 3 provides a detailed explanation of the proposed CGA-GWO algorithm;
Section 4 delves into a discussion of simulation results; finally,
Section 5 concludes the paper.
2. Related Work
This section outlines recent developments in the study of energy-aware cluster routing within sensor networks. In recent years, a trend has emerged towards incorporating hybrid approaches, including swarm-based metaheuristic optimization, aimed at lowering energy depletion and thereby extending network lifetime. Ajmi et al. [
11] proposed an energy-saving clustering approach known as the chicken-swarm-based genetic algorithm, which incorporates chicken-swarm-based optimization, cluster head selection, multi-weight clustering, a genetic algorithm, and cluster communication. Although this approach is notably effective in terms of energy efficacy, end-to-end latency, ratio of delivered packets, and network throughput, it might not be suitable for large-scale applications due to the possibility of communication delays.
Jia et al. [
12] proposed a clustering algorithm that utilizes ant colony optimization through adaptive chaos. To disrupt the pheromone along the route, the algorithm employs chaotic mapping while enhancing the transition probability using an adaptive technique. Using this approach, the local pheromone is updated and ants are sent out to adjust the present pheromone levels using a chaos factor. Nonetheless, this algorithm does not distribute transmission tasks effectively in scenarios involving a relationship between the cluster head and sink node. Wang et al. [
13] presented a cluster routing protocol that incorporates a self-organizing map neural network and firefly algorithm. Using the ant colony optimization algorithm, intercluster routing is used to select the next hop node by considering factors such as energy, distance, and node angle. The updating process of pheromones relies on the geometric coefficient of variation, and the routing path is improved by linking energy and distance.
Majeed et al. [
14] introduced a fuzzy-based genetic algorithm designed for cluster formation and cluster head selection, achieving fuzzy rule optimization and output value modification for membership in fuzzy logic. In addition, the ant colony process was employed to estimate the shortest path from each cluster head to the base node. A limitation of this approach lies in its focus on short routing, which leads to a suboptimal load balance in the network. Ram et al. [
15] introduced an effective approach for selecting cluster heads using the K-genetic algorithm. In this method, sensors are assembled into clusters based on their locations using k-means clustering, and a genetic algorithm is implemented to determine the most suitable cluster head. A trust-based firefly algorithm is applied to ensure secure and optimal routing.
Agrawal [
16] proposed a method for selecting cluster heads through the application of grey wolf swarm intelligence. The conventional GWO algorithm was tailored to the specific objective of selecting cluster heads in sensor networks, where the cluster formation function is determined by various factors including the balancing feature of the cluster heads, residual energy, average distance of each cluster, and sink distance. Joseph et al. [
17] proposed an enhanced chaotic GWO algorithm to enhance energy awareness in sensor networks, focusing on the selection of optimal cluster heads and routing paths. This approach aims to increase the percentage of packets delivered inside the sensor network while reducing node energy consumption.
Patra et al. [
18] introduced an energy-efficient clustering technique that employs a genetic algorithm. For shortest-path routing, they utilized ad hoc multipath routing by adapting grey wolf intelligence. Their approach identifies an optimal multipath route from paths generated during on-demand routing, where gray wolf intelligence is used to anticipate the optimal path. Singh et al. [
19] introduced an energy-saving routing algorithm that employs fuzzy GWO, aiming to minimize power consumption and achieve a balance in power usage among nodes in the sensor network. Additionally, the algorithm facilitates the reliable collection of link data by designated nodes and enhances throughput by mitigating traffic resulting from buffer occupancy in the nodes.
Gunjan et al. [
20] proposed a protocol based on a metaheuristic approach for clustering and routing in sensor networks, where a heuristic search is deployed to select cluster heads by considering three distinct fitness factors: the residual energy of each cluster head, the distance between each cluster head and the sink, and intercluster isolation. To direct data towards the base station, three additional fitness functions are applied targeting the residual energy of the next hops, distance to the next hop node from each cluster head, and total hop count. Liu et al. [
21] introduced the LEACH protocol based on a genetic algorithm comprising three phases for each round: preparation, setup, and data transmission. During the preparation phase, all nodes transmit messages indicating their candidacy statuses as cluster heads and share their locations with the base station. Using a genetic algorithm, the base station determines the optimal probability of each node becoming a cluster head. Subsequently, in the setup phase, the base station broadcasts a message containing the probability of all nodes forming clusters. The LEACH protocol was adopted for the setup and data transmission phases of each round of the protocol.
In all aforementioned studies, the genetic algorithm was used to encode clustering and routing processes within a single chromosome, resulting in the global estimation of energy. In addition, previous hops were not considered as parameters in the fitness functions for network load balancing. At present, there is a notable focus on leveraging the combined power of the chaos genetic algorithm (CGA) and grey wolf optimization (GWO) to enhance energy efficiency in sensor networks. This combination is gaining traction as a promising strategy to optimize energy consumption, extend network lifetime, and improve overall performance in sensor network deployments. The chaos genetic algorithm introduces chaotic dynamics into the genetic algorithm framework, promoting exploration and preventing premature convergence. This chaotic exploration enables CGA to effectively search for optimal solutions in complex and dynamic environments, making it well suited for addressing the challenges inherent in energy optimization in sensor networks.
On the other hand, grey wolf optimization mimics the social hierarchy and hunting behavior of grey wolves to optimize solutions in a distributed manner. GWO’s hierarchical search strategy allows for efficient exploration of the solution space, enabling the identification of energy-efficient configurations and resource allocation strategies within sensor networks. The synergy between CGA and GWO has led to significant strides in enhancing energy efficiency within sensor networks. This progress is evident across several avenues, notably through the development of energy-aware clustering mechanisms, routing protocols, and energy-harvesting optimization techniques. In contrast, the presently proposed CGA-GWO algorithm encodes clustering and routing in different chromosomes while using energy consumption and previous hops as parameters to evaluate the fitness function.
2.1. Chaotic Genetic Algorithm
The genetic algorithm, much like other evolutionary algorithms, is founded upon randomness. Consequently, local convergence and high tolerance for results owing to randomness represent primary drawbacks of the genetic algorithm [
22]. To enhance the performance of the conventional genetic algorithm, chaotic maps are employed in place of conventional random functions to generate random values. As a nonlinear phenomenon in nature, chaos can be used to mirror system complexity. Furthermore, the role of randomness introduced by chaotic dynamics resembles that of random variables. Optimization methods leverage chaotic systems to generate data with random values [
23].
Chaotic systems, such as Hennon and logistic maps, can substitute for randomness in the initial population, as well as the mutation, and crossover processes of the genetic algorithm. As the initial population generates random solutions, the randomness in the crossover creates new offspring. The randomness inherent to the mutations alters some genes in the offspring [
24]. Thus, the significant attributes of chaotic systems—namely pseudo-randomness and determinism—serve as compelling reasons for employing them in place of random processes within the chaotic genetic algorithm. This choice helps circumvent local convergence issues, thereby enhancing the traditional genetic algorithm’s performance [
25].
2.2. Grey Wolf Optimization
Mirjalili et al. [
26] introduced the GWO algorithm inspired by the stalking behavior of grey wolves in nature. Specifically, gray wolves can be classified as alpha, beta, delta, or omega. The GWO algorithm involves three steps: surrounding prey, hunting prey, and attacking prey.
2.2.1. Surrounding Prey
The surrounding behavior of grey wolves is mathematically represented by Equations (1) and (2):
where ‘t’ denotes the round number,
and
are vectors,
denotes the location of the prey,
is the location of the wolf, and
denotes the gap between the wolf and prey. The coefficient vectors
and
are computed using Equation (3):
where
and
are vectors with values in a range of [0, 1], and the value of
increases linearly from 0 to 2 throughout the iterations. The grey wolf updates its position by modifying
and
along with random vectors
and
and which aid the wolf in approaching the prey.
2.2.2. Hunting Prey
The hunting strategy of the GWO is orchestrated by optimal solutions provided by the alpha, beta, and delta wolves, enabling the prediction of prey locations. The locations of omega wolves are dynamically modified according to the coordinates of the alpha, beta, and delta wolves. The following equation can be used to represent this process:
where
denotes the wolf’s current position;
,
, and
denote the current locations of alpha, beta, and delta wolves, respectively;
,
and
represent the updated locations of alpha, beta, and delta wolves, respectively; and
,
and
are coefficient vectors. The absolute positions of wolves
,
and
are calculated using the following equations:
where
,
and
represent random vectors, and
t specifies the current iteration.
2.2.3. Attacking Prey
To execute an attack, the parameter is reduced and the values of vector are concurrently decreased by a within the interval of [−2a, 2a]. The parameter a gradually diminished from 2 to 0 in successive iterations. If the random values of fall within the interval of [−1, 1], the subsequent location of the searching agent will be the point lying between its present position and the location of the prey. If |U| < 1, the wolf is compelled to initiate an attack on the prey. Conversely, if |U| > 1, the wolves separate from each other in their pursuit of locating the prey. In the proposed algorithm, cluster heads are represented by grey wolves, with the base station serving as the prey.
4. Simulation and Results
The proposed chaotic genetic algorithm was evaluated using a MATLAB simulation under the assertion that the chaotic dynamics used in the algorithm are sufficient to achieve energy-efficient clustering. However, evaluating the efficacy of GWO-based routing involves assessing key parameters—including network lifetime, energy efficiency, throughput, and routing overhead—using a network simulation tool (Network Simulator 2.35). Although MATLAB is used for simulating the clustering process, the performance graphs were created using NS2.35. In NS2.35, various network scenarios are generated and obtain trace files containing crucial simulation events data, such as packet transmissions and receptions, among other metrics. These trace files were then formatted appropriately to ensure compatibility with Xgraph, a visualization tool within NS2.35. Xgraph offers a wide array of options for customizing plot appearance, including line styles, colors, and axis labels. Moreover, it allows for the plotting of multiple datasets on the same graph, facilitating easy comparison between different scenarios. After generating the plots, the visualized data are meticulously analyzed to derive meaningful insights into the performance of network simulation.
Table 1 presents the simulation configuration and parameters, along with their respective values or methodologies. The 100 × 100 m
2 network area size offers a realistic yet computationally manageable environment for sensor network simulation, accommodating diverse topologies. With 50–250 nodes, scalability across networks of varying sizes can be explored, from smaller setups where efficiency is less critical to larger ones. Placing the base station at (90,90) simplifies evaluation by centralizing data aggregation and transmission efficiency. A 2 m transmission range ensures ample connectivity without excessive interference or energy use. Initial node energy at 3 J provides a realistic starting point. Transmission and receiving energies (0.6 J and 0.2 J) are chosen based on theoretical estimates. A packet length of 4000 bits aligns with typical sensor network data sizes. A population size of 20 and 100 generations balances computational efficiency with thorough solution exploration. The elitist selection method maintains the best-performing solutions, aiding convergence towards optimal outcomes.
The efficacy of the proposed CGA-GWO algorithm was evaluated through a comparative experiment with existing schemes, including the low-energy adaptive clustering hierarchy genetic algorithm (LEACH-GA) [
21], fuzzy GWO [
19], and genetic-algorithm-based unequal clustering and routing (GA-UCR) [
20]. Performance metrics including the active node count, remaining network energy, packet transmission ratio, clustering overhead, and routing overhead were measured to evaluate each method’s efficacy. The performance measures are explained as follows:
Number of alive nodes: This metric measures the number of nodes in the network whose reserved energy has not yet depleted.
Average remaining energy: This metric indicates the average amount of energy still available among sensor nodes active in the network.
Percentage of packets received: This metric represents the ratio of packets successfully received by a node to the total number of packets it attempted to send within the network.
Clustering overhead: This metric refers to the additional control and communication costs associated with the establishment and maintenance of clusters within the network.
Routing overhead: This metric refers to the additional communication and computational costs incurred by routing protocols in managing and optimizing data transmission within the network.
The classic LEACH protocol presents several shortcomings that render it less suitable for comparison with the proposed CGA-GWO algorithm. These drawbacks include unequal cluster sizes, inefficiency with network size, limited support for multi-hop communication, and static network assumptions. Due to these inherent limitations, this research work does not include LEACH classic in the comparative analysis. Furthermore, according to the literature, LEACH-GA exhibits significant enhancements over classic LEACH in critical performance areas such as network lifetime, energy efficiency, throughput, and routing overhead. Moreover, existing research [
20,
27,
28] indicates that alternative algorithms such as LEACH-GA, Fuzzy-GWO, and GA-UCR offer notable improvements over LEACH classic in key performance metrics such as network lifetime, energy efficiency, throughput, and routing overhead. These algorithms have demonstrated enhanced effectiveness in managing energy consumption, extending network lifetime, optimizing data throughput, and reducing routing overhead when compared to the traditional LEACH approach.
Figure 5 depicts a plot of the number of live nodes in a 200-node network with respect to the number of rounds. According to the simulation results, the CGA-GWO algorithm exhibited more live nodes, surpassing GA-UCR, fuzzy GWO, and LEACH-GA by 15%, 30%, and 50%, respectively. In LEACH-GA, nodes are designated as cluster heads based on an optimized probability, leading to increased energy consumption. Moreover, the dead node count increased when the base station moved beyond the sensor field. In contrast, fuzzy GWO selects cluster heads according to residual energy, specifically targeting nodes with values that surpass the average remaining energy of all cluster heads. This involves the calculation of a fitness function that considers increased energy consumption, resulting in a higher number of dead nodes within the network. In GA-UCR, a genetic is applied for both cluster head selection and routing, combining the two schemes into a single chromosome. Consequently, energy consumption is a parameter in the fitness function calculation. As compared to other related algorithms, CGA-GWO shows a higher number of alive nodes as the number of iterations increases, as illustrated in
Figure 5. This improvement is attributed to the integration of chaotic features in the genetic algorithm and the strategic selection of nodes as cluster heads. CGA-GWO enhances search efficiency and increases the number of alive nodes by integrating chaotic elements into its genetic algorithm framework.
Figure 6 presents a plot of average remaining energy in a 200-node network over an increasing number of rounds. Here, CGA-GWO showed approximately 17%, 22%, and 30% more residual energy than GA-UCR, fuzzy-GWO, and LEACH-GA, respectively, by selecting nodes with the highest fitness values as cluster heads. The network can sustain operation for up to 700 rounds when the average remaining energy of the proposed algorithm CGA-GWO reaches 1 J by the 500th round. Specifically, the proposed algorithm determines suitable nodes for every generation of the population for energy-efficient routing. In addition, CGA-GWO selects chromosomes that consume less energy—i.e., have the maximum residual energy. These chromosomes are associated with the highest fitness values, establishing a direct proportionality between the fitness function and residual energy. Thus, the selection of cluster heads is facilitated by a chaotic genetic algorithm that relies on the residual energy of nodes in the network. This approach enhances the efficiency of cluster head selection and contributes to the overall energy optimization in the network.
Figure 7 shows the percentage of packets received by considering the total number of nodes in the network. For each algorithm, the percentage of packets received increased along with the number of nodes. Nonetheless, CGA-GWO outperformed LEACH-GA, fuzzy-GWO, and GA-UCR by 38%, 27%, and 14%, respectively, in terms of the percentage of packets received. The proposed CGA-GWO algorithm achieves a packet reception rate of 63%, which surpasses that of other related schemes. These results can be attributed to the fact that the GA-UCR algorithm neglects the existence of energy holes or hotspots formed during the routing process. In contrast, CGA-GWO maintains an increased number of delivered packets owing to its fitness function, which effectively minimizes packet loss during data transmission. To further reduce packet loss, CGA-GWO constructs an efficient path that does not contain dead nodes. Conversely, LEACH-GA and fuzzy GWO have high packet drop ratios stemming from inappropriate cluster head selection.
In
Figure 8, the clustering overhead of the CGA-GWO is compared with that of the three baselines by varying the number of nodes. CGA-GWO exhibited overhead reductions of 41%, 34%, and 11% compared with LEACH-GA, Fuzzy-GWO, and GA-UCR, respectively, as the proposed clustering fitness function selects high-residual energy nodes as cluster heads. Furthermore, the incorporation of chaotic features reduces the involvement of heavily loaded sensor nodes in the clustering process. In addition, the CGA-GWO algorithm employs an elitist selection method to select high-quality chromosomes, ensuring the transmission of elite individuals to subsequent generations. This ensures that higher-quality genes participate in each successive generation’s clustering process. In contrast, fuzzy GWO incurs a higher clustering overhead owing to repeated transmissions, whereas GA-UCR uses an intercluster separation fitness function to select cluster heads, resulting in increased overhead owing to the sizes of clusters. Ultimately, CGA-GWO employs a chaotic genetic algorithm to identify suitable cluster heads, minimizing the overhead associated with the clustering process.
In
Figure 9, the routing overheads of the different schemes are plotted for varying numbers of nodes, with CGA-GWO exhibiting a 47% reduction in overhead compared with the baselines. This is attributed to the consideration of factors such as the number of hops per path and residual energy per hop when selecting a routing path. By considering these factors, the fitness function facilitates the identification of an efficient route to the base station. The optimization aims to reduce the hop count along the path, resulting in a more efficient routing path compared with that determined by the baselines. In addition, the wolves’ arrangement and directional movements play crucial roles in determining the optimal routing solution. In GA-UCR, the classic genetic algorithm is employed to navigate data toward the base station given the NP-hard nature of the problem. In fuzzy GWO, nodes along the routing path remove the corresponding packets from their queues, leading to increased packet retransmissions and, consequently, a higher routing overhead.
The fitness function is specifically designed to incorporate these factors, aiding in the identification of efficient routes to the base station. This optimization strategy aims to minimize the hop count along the path, resulting in a superior routing path compared to other algorithms. Additionally, the arrangement of wolves and their directional movement within CGA-GWO’s genetic algorithm framework play a crucial role in determining the optimal solution for routing within the population. In contrast, GA-UCR utilizes a classical genetic algorithm to navigate data toward the base station, acknowledging the NP-Hard nature of the problem. Meanwhile, Fuzzy-GWO experiences increased routing overhead due to nodes along the routing path removing corresponding packets from their queues, leading to heightened packet retransmissions. In summary, CGA-GWO’s comprehensive approach to routing optimization, considering factors such as hop count and residual energy, coupled with its efficient genetic algorithm framework, contributes to its significant reduction in routing overhead compared to other schemes.