The Journal of Supercomputing (2025) 81:2

https://doi.org/10.1007/s11227-024-06474-2

Energy‑aware clustering method for cluster head selection

to increasing lifetime in wireless sensor network

Edris Alimohammadi1 · Sajad Haghzad Klidbary1 · Mohammad Javadian2

Accepted: 2 October 2024

© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature
2024

Abstract
Wireless sensor network consists of many tiny sensors that can be a powerful tool for
data collection in various environments. The optimal scenario for sensor networks is
for all nodes to reach the end of their energy, together or through regular schedul-
ing, to maximize the lifetime of network. Studies have shown that clustering helps
conserve the limited energy resources of sensors. In this paper, an ink drop spread is
used for clustering. Initially, ink drops are spread using the ink drop spread operator
with a weight proportional to the energy of each node, and clustering as well as rout-
ing is performed based on it. Clustering is performed dynamically; in each round,
clustering is redone, cluster heads are selected, and nodes with very low energy
are marked as dead. Our proposed algorithm, EACM, is compared with the popu-
lar algorithms LEACH, PEGASIS, LEACH_EX, and the latest algorithms EECPK-
means, RaCH, and C3HA. The proposed algorithm demonstrates an improvement in
clustering quality, a reduction in node energy consumption, and an overall increase
in the lifetime of network. On average, the proposed algorithm results in more active
nodes within the network, with the remaining energy being at least 17% higher than
that of the best existing algorithms.

Keywords Wireless sensor network (WSN) · Active learning method · Density-

based clustering · Energy optimization · LEACH

1  Introduction

In today’s technological era, there is a prevalent belief that everything should be

automated. Constant monitoring of the surrounding environment is essential,
necessitating robust tools to collect, process, and deliver vast amounts of data and

* Sajad Haghzad Klidbary

[email protected]; [email protected]
1
Department of Electrical and Computer Engineering, University of Zanjan, Zanjan, Iran
2
School of Electrical Engineering, Shahid Beheshti University, Tehran, Iran

Vol.:(0123456789)
 2 Page 2 of 23 E. Alimohammadi et al.

information to users [1–3]. Furthermore, life without wireless communications is

unimaginable. The advancement of CMOS technology and the creation of smaller
and smaller circuits have made wireless circuits possible in most electronic devices
today. This progress has also led to the development of microsensors and has given
rise to the idea of creating and expanding networks known as WSNs.
WSNs consist of numerous small sensors that can be powerful tools for collect-
ing data in various environments. Each sensor node consists of a sensor module,
a processing module, a communication module, and a battery that works together
to gather information. The collected data from each sensor are wirelessly transmit-
ted to a central station and presented to the end user. The main goal of WSNs is to
monitor and control weather, physical, or chemical conditions in a specific range of
environments [4–6].
In a WSN, sensor nodes are widely distributed in a specific environment and peri-
odically collect sensed data from the environment. These sensors perform minimal
processing on the data and then send the resulting information to the central sta-
tion. The network model has the following features: (1) All nodes have the same
capabilities in terms of sensing, processing, and communication. (2) All nodes have
energy constraints and have the same amount of energy. (3) Although sensor nodes
have energy constraints, they all have the same capabilities. (4) Nodes are randomly
deployed in the network and become stationary after deployment. (5) The base sta-
tion (BS) can be stationary or mobile and has no energy constraints. (6) All nodes
have enough energy to process various protocols and process signals. (7) Charging
or replacing the batteries of sensor nodes is not possible. (8) Cluster heads commu-
nicate their initial information to the sink through signaling.
A sensor node consumes a significant portion of its energy in data transmission
and neighbor discovery. The overall lifetime of a WSN heavily depends on how
neighboring nodes are set up and how data are routed to the destination. Therefore,
efficient energy utilization of sensor nodes can prolong the lifetime of WSNs. The
lifetime of WSNs depends on the lifetime of the sensors themselves, which also
have limitations in terms of power, computations, memory capacity, and storage.
The ideal state in sensor networks should be such that the energy of all nodes is
exhausted together or through regular scheduling. Thus, effective energy utilization
of sensor nodes can result in longer lifetimes for WSNs. Most of the energy of nodes
is spent on data collection from the environment and data transmission. The most
significant challenge in WSNs is managing the energy consumption of the sensors,
which directly impacts the lifetime of network. Various protocols and algorithms
have been proposed to regulate energy consumption within specific criteria [7–11].
In a sensor node, the maximum energy is consumed in receiving and sending
data, as well as processing the received data. The energy consumed in packet trans-
mission has a direct relationship with the transmission distance, meaning that the
longer the distance a sensor node sends a packet, the more energy it will consume.
Additionally, the energy consumption is directly related to the packet size, mean-
ing that larger packets require more energy to be sent. There are two methods for
data transmission, one being single-hop and the other being multi-hop. In the sin-
gle-hop method, each sensor directly sends its information to the central station. In
this method, a large amount of energy is consumed due to the long distance of the
Energy‑aware clustering method for cluster head selection… Page 3 of 23 2

sensors from the center. On the other hand, the multi-hop approach reduces com-
munication distances with the help of intermediate sensors and can prolong the life-
time of network. It can be said that multi-hop communication is more valuable and
cost-effective in such networks compared to single-hop communication. However,
in multi-hop communication, more energy is consumed by the nodes to establish
connections with other sensors, which results in high-energy consumption in the
sensors.
WSNs have thousands of nodes that need proper energy management [12–14].
Various methods have been proposed to reduce energy consumption in wireless sen-
sor networks, which can significantly reduce energy consumption by performing
network operations in a way that minimizes energy consumption. In all conducted
research, it can be stated that routing protocols and clustering have occupied a sig-
nificant amount of research. Clustering of sensors plays an essential role in reducing
energy consumption [15–17].
Therefore, using clustering methods reduces the size of the routing table, saves
consumed bandwidth, and facilitates network management. Clustering is an effective
method for providing better data aggregation and scalability for large WSNs. This
critical role of clustering in reducing energy consumption is based on the proper
selection of cluster heads (CHs). The selection of CH is the most critical factor in
clustering and helps save energy depending on the distance of transmitted packets.
Therefore, the CH node incurs a high additional burden due to activities such as data
aggregation and sending data toward the sink. CH nodes are more prone to prema-
ture death due to high-energy consumption than others. For this reason, selecting a
CH is considered one of the main challenges in this field. Choosing a CH is a mul-
tivariable decision, so selecting the CH node is one of the most complex challenges
known as an NP-HARD problem. Choosing a CH node using mathematical methods
is also possible, but it is very time-consuming in large networks. In clustered WSNs,
the BS and the CHs need to be placed correctly for proper communication. Location
differences facilitate network lifetime and data transmission efficiently. Determining
suitable positions for sensors in the network is still an ongoing topic, and there is
much attention for development [18–22]. Since all nodes in a cluster must send their
information to the CH, and energy consumption is directly related to distance, it is
better for the CH to be placed as close as possible to the center of cluster.
Fuzzy logic is considered one of the most effective techniques for modeling
uncertainty, and uncertainty is an inseparable part of our surrounding world. Fuzzy
clustering has many advantages over crisp clustering, the most important being the
consideration of overlap and uncertainty for actual data. Therefore, due to the effi-
ciency of fuzzy logic in many applications, it has been used to solve the main chal-
lenge of CH selection.
One of the innovations of this paper is the use of ink drop spread in WSNs. The
main part of the proposed algorithm is based on the IDS operator, where ink drop is
spread for each node in the sensor network. These ink drops will overlap with each
other and eventually create patterns that will help in selecting the CHs with extreme
values. As a result, an algorithm with low computational complexity and high speed
has been achieved to increase the lifetime of WSNs. The innovations of this paper
can be summarized as follows:
 2 Page 4 of 23 E. Alimohammadi et al.

(1) Fuzzy clustering has been used to find optimal clusters, resulting in optimal CH
selection.
(2) Ink drop spread (IDS) operator has been used to find CHs based on the energy
of nodes.
(3) The CH selection algorithm is efficient and contributes to a longer lifetime.
(4) The proposed algorithm has a longer lifetime and lifespan (in terms of the death
of half of the nodes and the total energy of the network) compared to other
standard clustering algorithms in WSNs.
(5) The algorithm does not require advanced communication with sensor nodes, and
all nodes are simulated as regular sensor nodes.
(6) With the appropriate selection of CH locations, the number of sent packets in
the proposed algorithm is optimal, resulting in lower energy consumption.
(7) Many of these previous works require the pre-defined number of clusters to be
specified as a parameter. This is problematic because we usually do not know
the number of clusters in clustering tasks.

The rest of the paper is organized as follows: Section 2 discusses the related
works. In Sect. 3, the proposed algorithm is presented. Section 4 presents the energy
and network model used to analyze and evaluate our proposed algorithm. Lastly,
Sect. 5 includes an overview, conclusion, and suggestions for future work.

2 Definition and previous work

WSNs must transmit the data from the environment to the recipient through a sink.
The energy consumption in data transmission is directly related to the distance
between the sender and the recipient, as well as the size of the data. In clustering,
sensors are generally divided into two categories: regular sensors and CH. Regular
sensors need to send data to their CH, while the CH needs to forward it to the sink or
the next CH. Determining the CHs plays a crucial role in data transmission, leading
to the development of numerous algorithms in the literature. These algorithms will
be discussed further in the study. Let us assume that sensors in the environment are
represented by the following relationship.
S = {a1 , a2 , a3 , … , an } (1)
where ai , is network nodes that the set of randomly deployed. Clustering algorithms
should decide which sensors should be included in which clusters and which sen-
sors should be in charge of CHs. CHs collect data from regular sensors inside the
cluster; therefore, they consume more energy than regular sensors. To overcome
this problem, correctly determining CHs is vital. Various methods have been used
to reduce energy consumption in WSNs, among which clustering-based routing
algorithms have been proposed. Clustering methods reduce data transmission with
less energy consumption. Therefore, CH selection methods are one of the essential
factors affecting network energy. Clustering is the process of organizing elements
into groups that are similar to each other, and the goal of clustering is to quickly
Energy‑aware clustering method for cluster head selection… Page 5 of 23 2

and accurately achieve related information and identify logical connections between
them. In these protocols, nodes within clusters are organized so that nodes with
high energy can be used for processing and data transmission, while nodes with low
energy can be used for target detection. The data of each cluster are only sent to
a CH node. The CH selection method considers different criteria in each method.
In most clustering methods, the main goal is to achieve a uniform distribution of
energy consumption among all nodes. Clustering protocols are commonly classified
into two categories: clustering in homogeneous or heterogeneous networks, and cen-
tralized and distributed clustering algorithms.
In homogeneous Ad-Hoc networks, all nodes have the same specifications and
hardware capabilities, so every node can act as a CH to achieve better load balancing
and more integrated energy consumption, and the exchange of roles among nodes
can occur. In heterogeneous Ad-Hoc networks, the nodes have different hardware
configurations, and each node has different specifications, resources, and policies,
and nodes cannot provide similar services. In a network using a centralized routing
protocol, a central processing device running on a central node collects information
available on each link in this network. Then, this processing device uses the col-
lected information to calculate routing tables for other nodes. This type of routing
protocol uses a centralized database in the central node for these calculations. In
other words, the routing table is maintained in a central node and used when other
nodes need to make routing decisions. Under a distributed routing protocol, each
device in the network is responsible for making routing decisions. There are two
types of distributed protocols, namely distance-vector and link-state protocols, and
these subsets form two classes of protocols that are widely used today. The distance-
vector protocol shares information about the destination and cost among nodes at
regular intervals or as needed. The link-state protocol disseminates link status infor-
mation throughout the network to allow each node to form a map of the network.
Various papers have been studied in the following analysis.

2.1 Related works

One of the most important studies is the LEACH protocol, developed by Heinzel-
man. This protocol is one of the most well-known hierarchical protocols for wireless
sensor networks [18]. In this protocol, the selection of CHs is performed using a
probabilistic function, and LEACH assumes that each node has the required radio
capability to transmit to the main base or the nearest CH. The main equation of the
algorithm is as follows:
{ P
( ( )) if n ∈ G
1−P rmod P1
T(n) = (2)
0 O.W,

P represents the probability of flipping heads, r represents the current round num-
ber, and G represents the set of nodes that have not flipped heads in the last 1/P
rounds. This protocol has been widely adopted due to its advantages and high effi-
ciency, as depicted in Fig. 1.
 2 Page 6 of 23 E. Alimohammadi et al.

Leach-derived protocols

Protocols based on CH selection and

Protocols based on CH selection Protocols based on data transmission
data transmission

Multi/Single-
E- Leach Single-step Multi-step
step

BRE- Leach Leach- F A- Leach CELL- Leach

Residual Energy
Concentration Mobility Distance EDMHT V- Leach Leach- TL O- Leach
energy efficiency

W- Leach L- Leach MH- Leach

Leach- B VH- Leach
Leach- M I - Leach
Leach- C MHT- Leach
WD- Leach
Leach- T
I- Leach Leach- ME EADCR
IMHT- Leach
TB- Leach
EN- Leach DMHT- Leach

Fig. 1  The types of protocols derived from LEACH protocol (Three main categories: based on CH selec-
tion, based on CH selection and data transmission, and based on data transmission)

The LEACH_E [23] protocol, similar to the LEACH protocol, does not con-
sider the same initial energy level for all clusters at the beginning. This proto-
col assumes that sensors have general information about the remaining energy
of other nodes. Additionally, this algorithm determines the required number of
clusters to be changeable and dependent on parameters such as the square root of
the sum of sensor nodes, in order to reduce energy consumption.
In the LEACH_C [24] protocol, which is based on the LEACH protocol, there
is an improvement in cluster selection and sensor selection compared to the
LEACH protocol. The LEACH-C protocol provides the possibility of selecting a
two-layer CH. This way, the use of a central control algorithm in creating better
clustering is more effective and prevents unnecessary energy consumption. In the
framework of the LEACH-C protocol, all nodes in a sensor network send their
current position and energy level to the sink, and these data are used to determine
the threshold. Nodes with values below this threshold cannot be selected as CH.
LEACH_EX [25] protocol has a better performance compared to the two pro-
tocols: LEACH and LEACH_E. In this protocol, simplification and improvement
of the threshold calculation formula have been done, which has resulted in better
performance. For the sake of brevity, the explanation of the remaining matters
related to the LEACH family has been avoided.
The PEGASIS [26] protocol is an evolved form of the LEACH protocol. In this
protocol, instead of forming different clusters, a chain connection is established
between all sensors in the network, so that each sensor is in communication with
its neighbors for sending and receiving data, and only one sensor in the entire
network is selected as the data transmitter to the base station. The collected data
from the sensors are sent to the neighboring sensors. Each sensor that receives
data from its neighbor combines it with its data and sends it to the next sensor.
This process continues until the data reach the sensor communicating with the
BS. It is worth mentioning that other algorithms, such as H-PEGASIS [27] and
Multi-Chain-PEGASIS [28], have also been proposed based on PEGASIS.
Energy‑aware clustering method for cluster head selection… Page 7 of 23 2

EECPK-means [29] is a protocol that utilizes the K-Means algorithm for

cluster identification and is used for homogeneous networks. In this protocol, a
threshold value is determined for selecting a CH, and in each round, if the energy
value of a node exceeds that threshold, that node can become a CH. While the
node death is delayed in this algorithm, the network lifetime is low. A new algo-
rithm has been developed in [30] to increase the lifetime based on the LEACH
algorithm and to reduce energy consumption; the CH selection criteria are based
on the distance to the BS.
The RaCH [31] algorithm has been proposed for a homogeneous WSN. In this
algorithm, virtual circular regions are identified for selecting CHs. The circles
start from the BS and progress toward the end of network. The region is divided
into four quadrants, and the distance between the BS and the circles plays an
essential role in identifying these quadrants. In each quadrant, the CH is selected
based on specific considerations.
The C3HA [32] algorithm is proposed as a new clustering method for WSNs.
It is designed for cluster-based sensor networks and utilizes the K-means algo-
rithm to determine the network clusters of interest. Unlike existing algorithms,
C3HA requires dual clustering. This means that first, the network is clustered
using K-means, and then, a specific subset of nodes is determined for each clus-
ter. This subset is called the cluster centered (CC), and the nodes in CC have
priority when selecting CH. The set of nodes outside CC, referred to as the com-
plement of CC or the out-cluster center, can only become CHs when all the nodes
in CC are dead. The problems of previously proposed algorithms can be listed as
follows:

(1) The selection of cluster heads in each round cannot provide a suitable conver-
gence guarantee due to the random selection of CHs. This is because the cluster
heads can be closed to each other if they are randomly selected. As a result,
energy consumption increases.
(2) The probability of selecting a node with lower energy or further distance than
selecting a node with higher energy and closer distance is the same. Therefore, if
these nodes are chosen as CHs, they will deplete their energy faster than others,
resulting in a shorter cluster lifetime.
(3) In the selection of CHs, the remaining energy of the nodes is ignored. However,
clusters have higher traffic costs than non-cluster nodes due to their role in col-
lecting and transmitting data from all cluster members to the sink.
(4) The use of single-hop communication by the CH node to establish communica-
tion with the sink node makes the protocol unusable for large networks.
(5) Duplicate data collisions can reduce network efficiency in many cases.

We have proposed a new algorithm to solve the problems and reduce energy
consumption. In the proposed algorithm, for each round, CHs are selected based
on their energy levels and neighbors in the locations close to the cluster center.
This selection reduces the energy distance between CHs and sensors. The details
of the proposed algorithm are provided below.
 2 Page 8 of 23 E. Alimohammadi et al.

Fig. 2  The functionality of the ALM algorithm in dealing with a multi-input single-output (MISO) sys-
tem

y
ymax

Y Function of Narrow Path

( 4,13)

(10, 9) Spread at point x= Xt*

*
Y
ymin

(13, 6)

X min X max Xt* X

(a) (b) (c)

Fig. 3  a The hypothetical training data, b ink drop spread operator, and c feature extraction (NP and SP
features)

3 Proposed algorithm

Before introducing the proposed algorithm, the ink drop spread (IDS) operator
used in ALM, which is the basis of the proposed algorithm, is examined.

3.1 Active learning method (ALM)

The ALM is based on the principle that humans break down complex problems
into smaller ones when dealing with them. This method has various applications
such as control, robotics, modeling, classification, and clustering [33–39], and its
functionality is illustrated in Fig. 2.
The core of this method is based on the IDS operator, where for each train-
ing data, an ink drop is dispersed. These drops overlap, resulting in patterns
from which two main characteristics, the narrow path (NP) and spread (SP), are
extracted from each sub-system. The final output is a weighted sum (WS) of the
output features from each sub-system. A representation of the ink drops and the
pattern created by their overlapping is shown in Fig. 3.
Finally, after the spreading of ink drops on two-dimensional surfaces and the
extraction of NP and SP features, a weighted summation of these feature values is
used to find the desired output.
Energy‑aware clustering method for cluster head selection… Page 9 of 23 2

3.2 The proposed algorithm (EACM)

As mentioned, in WSNs, CH nodes bear significantly higher extra load due to

specific activities such as data aggregation and data transmission toward the sink.
Therefore, selecting CHs poses one of the main challenges in this field. CH selec-
tion is known as an NP-HARD problem. In this paper, we aim to find appropri-
ate CHs based on proposing a fuzzy clustering algorithm. Considering that the
objective of the proposed algorithm is to increase the network lifetime, in each
round, selecting the best node as the CH will significantly contribute to the life-
time of network. Distributed nodes in a network have varying degrees of impor-
tance depending on their position, energy, and distance to the BS. Hence, each of
these nodes can be assigned a degree of importance, and this importance degree
is applied within the amplitude of the ink drop spread (proportional coefficient).
Finally, the extreme values (maxima obtained) of these patterns are used as clus-
ter representatives. This process is executed based on the system policy until
the energy of the first node or the total energy of the network is depleted. The
flowchart of the proposed algorithm is depicted in Fig. 4. The clustering-related
section is depicted in the figure, and other sections, such as node deployment,
parameter value calculations, information transmission, and energy computations,
are the same in all WSN methods. The steps related to the proposed algorithm
will be explained as follows:
First step: Quantizing the input data is the first step of the algorithm. In order to
achieve faster simulation time and require less hardware for implementation, the fea-
ture range should be quantized into appropriate levels ( Rs) (where D = 2 represents
the number of system features).
( ) ( )
x − xmin × Rs [ ) y − ymin × Rs [ )
Xq = + 1, x ∈ xmin , xmax , Yq = + 1, y ∈ ymin , ymax ,
xmax − xmin ymax − ymin
(3)
Xq , Yq ∈ {1, 2, ..., Rs} (4)
Second step: The parameters that must be determined in the initial step include
the ink radius (Ir ) and the threshold value (Th). These two parameters are essential
for the algorithm to find the desired or optimal number of clusters. The Ir parameter
determines the influence of a point on its neighborhood, while Th can be used for
noise removal and determining cluster boundaries, affecting clustering accuracy and
learning speed. When the data size is small, the size of the ink drop pattern should
be large. If the data size is sufficiently large, a smaller ink drop pattern increases
the learning speed and improves clustering accuracy. A smaller ink drop pattern
requires fewer addition operations, reducing computational complexity. Further-
more, as the size of the ink drop decreases, the clusters become more separated,
resulting in increased accuracy. It is worth mentioning that for determining the algo-
rithm parameters, if we have an initial estimate of the number of clusters, it can
help us find suitable values for the ink radius and threshold through trial and error.
Otherwise, we can utilize an evaluation metric to find optimal parameter values. If
a suitable evaluation metric is chosen as the cost function, exploratory optimization
 2 Page 10 of 23 E. Alimohammadi et al.

Fig. 4  Flow diagram of the proposed algorithm with six steps in the clustering section (The three main
parts are shown in the figure: quantization, spreading, and finding clusters)

algorithms such as genetic algorithms can be used to determine the ink radius and
threshold parameters automatically.
Third step: In this step, similar to the initial version of ALM, each training data
point is represented by a fuzzy membership function that considers its uncertainty.
This representation is commonly known as the ink drop diffusion. During the diffu-
sion process, the ink drops merge, eventually forming interconnected clusters. The
Gaussian function is used as the following relationship. Finally, the range of varia-
tions can be normalized between 0 and 1.
( 2
x + y2
)
h(x, y) = exp − (5)
2𝜎 2

Fourth step: To determine the number of clusters, we need to define an opera-

tor on the IDS space. Continuous component labeling is used for this purpose,
which is one of the commonly used algorithms in graph theory and image process-
ing. This algorithm effectively separates dependent or connected components from
Energy‑aware clustering method for cluster head selection… Page 11 of 23 2

Fig. 5  a Spreading of ink drop for each node, b proposed window to finding connected components and
labeling, and c obtained clusters

Fig. 6  a Existing unlabeled data and b Ink drop respreads and finding new clusters for unlabeled data

other parts. By moving a window over the IDS pages, this algorithm assigns the
same label to adjacent cells and ultimately assigns them the same cluster number. A
graphical representation of this process is shown in Fig. 5.
Fifth step: Finally, with having zones or areas associated with each cluster, labe-
ling the primary data becomes more accessible, and each data point will have the
label of the cluster it falls into if it is within the defined boundaries of the cluster. If
some data points do not fall within a specific boundary (like Fig. 6), in this case, one
can distribute a drop again and compare the overlapping areas to determine the label
of that data point or utilize the KNN method to find the label.
Sixth step: By applying fuzzy clustering, the degree of membership is not limited
to only 0 or 1, and each cluster is assigned a separate IDS page, with relevant data
being allowed to be disseminated within that cluster, and ultimately, considering all
the data, each data point will be assigned a degree of membership in all clusters.
 2 Page 12 of 23 E. Alimohammadi et al.

Finally, it can be stated, the proposed algorithm has S(N) = O(R2s ) space com-
plexity, and T(N) = O(N × Ir2 ) time complexity. Rs is quantization level, Ir is ink
radius, and N is the number of nodes.

4 Performance analysis

All simulations have been performed on hardware with a Core i7, 2.4 GHz proces-
sor, and 16 GB RAM. The evaluation of the proposed algorithm includes criteria
such as data transmission and reception, the death time of the first node, the death
time of half of the nodes, and network energy. These criteria have been compared
with algorithms such as LEACH, PEGASIS, LEACH_EX, EECPK-means, RaCH,
and C3HA.

4.1 Energy model

The energy consumption model in WSNs should be such that the energy consump-
tion is equal throughout the network and the lifetime of network is extended. The
energy consumption for data transmission operations includes energy from elec-
tronic circuits and transmission amplifiers, while the energy consumption for data
reception operations only includes energy from electronic circuits. Therefore, data
transmission operations consume more energy than data reception operations. Fur-
thermore, the amount of energy consumed in wireless data exchange is higher com-
pared to the case where sensory or memory operations are performed in the sensor.
In the considered model, all the energies consumed by sensor nodes are accounted
for data exchange, disregarding the energy consumed for other operations. The radio
model is depicted in Fig. 7.
The relations will be as follows:

ETX (k, d) = Eelec ∗ k + (Eamp ∗ k ∗ d2 ) (6)

( )

ERX = Eelec ∗ k (7)

The energy consumption model in the above relation is represented by ERX for
the receiving node and ETX for the transmitting node. Eelec is the energy required to
transmit and receive one bit of information, regardless of the distance. Eamp is the

Fig. 7  Energy consumption model as considered (first order radio model)

Energy‑aware clustering method for cluster head selection… Page 13 of 23 2

energy required to amplify the transmitted signal over the desired distance. k repre-
sents the message length, and d represents the distance to the receiving node.

4.2 Lifetime of network

As mentioned, the goal of network energy management is to increase the lifetime

of network. The lifetime of network is essentially the duration until the total energy
of the network is depleted, or in other words, the duration until the energy of the
last node is exhausted and the entire network becomes unresponsive. In most cases,
this duration is normalized by dividing it by its maximum value and, therefore, falls
within the range of 0 to 1. Different references define the network lifetime based
on various criteria, including: (1) The death time of the first node: the round num-
ber in which the first node becomes non-functional due to energy depletion. (2) The
death time of half of the nodes: the round number in which half of the nodes in the
network become non-functional due to energy depletion. (3) The death time of the
last node: the round number in which the last node in the network becomes non-
functional due to energy depletion. (4) Total network energy: the sum of the energy
available in the network nodes.

4.3 Results

Two different scenarios have been examined to evaluate the performance of the pro-
posed algorithm compared to other algorithms. A simulation environment of size
100 by 100 is considered, wherein one scenario, 100 sensors are randomly distrib-
uted, and in the other scenario, 200 sensors are randomly distributed. A hypothetical
point is assumed as the location of the BS, as shown in Fig. 8. The proposed algo-
rithm and other existing algorithms have been investigated and evaluated in these
environments.
The simulation parameters related to algorithm simulations in the two proposed
scenarios are listed in Table 1. These models are characterized by: random distribu-
tion of nodes, static position of nodes, stationary sink node, and the depletion of the
battery lead to the death of the sensor.

4.3.1 First scenario

In this scenario, 100 nodes are distributed in the environment. The related Figs. 9,
10, 11 and 12 have been provided for the total network energy, the number of sent
packets, and the number of dead nodes. Additionally, the results have been presented
quantitatively in Table 2.
Figure 9 shows the energy consumption of network versus rounds in all algo-
rithms. The total network energy (sum of energy for all nodes) is 30 J. As can be
seen from the figure, the proposed algorithm has the best performance. The main
reason for that is using the energy coefficient in the ink drop spreading amplitude
to find CH. Therefore, unlike other algorithms, there is no random selection in the
 2 Page 14 of 23 E. Alimohammadi et al.

(a) (b)

Fig. 8  a Hypothetical environment with 100 nodes and b hypothetical environment with 200 nodes and
fixed location of BS in both environments

Table 1  Simulation parameters used in two proposed scenarios

# Definition Symbol Scenario 1 Scenario 2

1 Number of nodes N 100 200

2 Sigma 𝜎 0.15 0.2
3 Threshold TH 0.25 0.15
4 Location of node – Random Random
5 Size of area M×M 100 × 100 m2
6 BS location – (50,175)
7 Initial energy E 0.3J
8 Maximum round Rmax 1000
9 Data packet size K 500 Bytes
10 Transmission and Reception Energy ETX and ERX 50 nJ∕bit
11 Cluster head probability p 0.1

selection of the cluster head, and this has improved the efficiency of the algorithm.
EACM is 17.18% more efficient than the best algorithm on average.
The amount of packet sent to the sink node is essential for network performance.
With Fig. 10, the proposed algorithm sends more packets than others. Due to in the
proposed algorithm, the selection of cluster heads is intelligent, and the distance
between the cluster heads and other nodes are minimal, so the packets are sent with
the lowest energy and the largest number. Also, the random creation of the network
in the environment, the distances of the nodes are different, and the number of sent
packets depends on the distances of the nodes.
Energy‑aware clustering method for cluster head selection… Page 15 of 23 2

Fig. 9  Energy consumption of network versus rounds in all algorithms

Fig. 10  a Number of packets sent versus rounds in all algorithms and b zooming in details

Figure 11 shows the dead nodes versus rounds in all algorithms. According to
the figure, in the proposed algorithm, the death of the first node occurred faster
than the rest of the algorithms, but through time, the number of dead nodes in
the proposed algorithm has a much lower growth rate than other algorithms.
The selection of nodes according to the energy causes the early death of dis-
tant nodes and this causes the death of the first nodes to occur earlier in the
proposed algorithm. From the figure, it is observed that our proposed EACM is
10% better than C3HA, 11% better than, 13% better than ECPK-means, 15% bet-
ter than LEACH_EX, 18% better than PEGASIS, and 21% better than LEACH
with respect to the parameter LND. Figure 12 compares criteria FND, HND, and
 2 Page 16 of 23 E. Alimohammadi et al.

Fig. 11  Dead nodes versus rounds in all algorithms

(a) (b)

Fig. 12  Comparison of rounds in all algorithms: a in each criterion and b sum of all node

LND for all algorithms. The sum of the rounds related to each algorithm indi-
cates the better performance of the proposed algorithm.
The summary of the results deduced from the figures is given in Table 2. The
results are presented as an average of 100 algorithm runs.
Energy‑aware clustering method for cluster head selection… Page 17 of 23 2

Table 2  The results of first scenario (regarding the number of rounds for all algorithms)
Algorithm First node dead Half node dead Last node dead Energy
(FND) round (HND) round (LND) round consumption(EC)
Joul

LEACH 70 390 552 3.14

PEGASIS 92 394 567 4.17
LEACH_EX 147 456 581 5.59
EECPK-means 72 448 592 12.08
RaCH 87 466 600 14.20
C3HA 237 468 605 14.84
EACM 71 520 670 17.39

4.3.2 Second scenario

In this scenario, 200 nodes are distributed in the environment. The related Figs. 13,
14, 15, and 16 have been provided for the total network energy, the number of sent
packets, and the number of dead nodes. Additionally, the results have been presented
quantitatively in Table 3.
Figure 13 shows the energy consumption of network versus rounds in all algo-
rithms. The total network energy (sum of energy for all nodes) is 60 J. As can be
seen from the figure, the proposed algorithm has the best performance. The main
reason for that is using the energy coefficient in the ink drop spreading amplitude
to find CH. Therefore, unlike other algorithms, there is no random selection in the

Fig. 13  Energy consumption of network versus rounds in all algorithms

 2 Page 18 of 23 E. Alimohammadi et al.

Fig. 14  a Number of packets sent vs. rounds in all algorithms and b zooming in details

Fig. 15  Dead nodes versus rounds in all algorithms

selection of the cluster head, and this has improved the efficiency of the algorithm.
EACM is 4.86% more efficient than the best algorithm on average.
As we know, the amount of packet sent to the sink node is important for network
performance. With Fig. 14, the proposed algorithm sends more packets than others.
Due to in the proposed algorithm, the selection of cluster heads is intelligent, and
the distance between the cluster heads and other nodes is minimal, so the packets are
sent with the lowest energy and the largest number. Also, the random creation of the
network in the environment, the distances of the nodes are different, and the number
of sent packets depends on the distances of the nodes.
Figure 15 shows the dead nodes versus rounds in all algorithms. According to
the figure, in the proposed algorithm, the death of the first node occurred faster than
Energy‑aware clustering method for cluster head selection… Page 19 of 23 2

(a) (b)

Fig. 16  Comparison of rounds in all algorithms: a in each criterion and b sum of all nodes

Table 3  The results of second scenario (regarding the number of rounds for all algorithms)
Algorithm First node dead Half node dead Last node dead Energy
(FND) round (HND) round (LND) round consumption
(EC)
Joul

LEACH 75 457 756 0.959

PEGASIS 60 461 769 6.420
LEACH_EX 82 539 777 10.25
EECPK-means 83 540 783 16.61
RaCH 86 627 804 21.27
C3HA 178 629 813 22.39
EACM 77 737 909 23.48

the rest of the algorithms, but through time, the number of dead nodes in the pro-
posed algorithm has a much lower growth rate than other algorithms. The selection
of nodes according to the energy causes the early death of distant nodes, and this
causes the death of the first nodes to occur earlier in the proposed algorithm. From
the figure, it is observed that our proposed EACM is 11% better than C3HA, 13%
better than RaCH, 16% better than ECPK-means, 17% better than LEACH_EX, 18%
better than PEGASIS, and 20% better than LEACH with respect to the parameter
LND. Figure 16 compares criteria FND, HND, and LND for all algorithms. The
sum of the rounds related to each algorithm indicates the better performance of the
proposed algorithm.
The summary of the results deduced from the figures is given in Table 3. The
results are presented as an average of 100 algorithm runs.
Simulation results show that in the proposed algorithm, the death of the first node
occurs earlier than in some algorithms. However, the energy stability is better and
higher than other algorithms (meaning that the overall energy decrease occurs with a
 2 Page 20 of 23 E. Alimohammadi et al.

delay). Also, the proposed algorithm performs better in other criteria. In the energy
consumption calculation, the number of packets, clusters, and the distance between
the nodes are the effective factors. It is worth mentioning that in the developed ver-
sions of the LEACH algorithm, a threshold criterion is set. If a node is chosen as
a cluster head and its energy are less than a threshold, it will be disregarded, and
other nodes will be selected. This has caused the relative stability of these algo-
rithms compared to the main LEACH algorithm. The proposed algorithm can effec-
tively reduce the speed of node death and increase the lifetime of network, along
with balanced energy consumption. One of the reasons for the proposed long-term
of algorithm performance is that, in this algorithm, further nodes die earlier, and due
to reduced energy consumption distance, it decreases.
One of the disadvantages of the proposed algorithm is that if the further nodes
die, the coverage of these areas will be lost. If we are interested in WSNs, this cover-
age is always continuous. Therefore, to increase the lifetime of network, efforts are
made to have a uniform distribution of nodes in the network, so that the time inter-
val between the death of the first node and the last node is minimized.

5 Conclusion

Advancements in industry and technology have led to the development of wireless

communication, especially the use of sensors in WSNs. WSNs consist of a collec-
tion of nodes that are densely deployed in an environment and work together to per-
form a common task. Time, energy management, and control are among the most
important challenges in these types of networks. Routing protocols transfer data
from the source node to the destination nodes. Due to the lack of a centralized struc-
ture and limited resources, especially energy constraints in WSNs, an appropriate
routing protocol is necessary for these networks. In this paper, CH selection was
performed using the IDS operator method, taking into account the energy of the
nodes. Therefore, we overcome the limits of the random selection of CHs selection
algorithms through a new algorithm that computes the best node to be selected as
CH. The simulation of the proposed algorithm and the analysis of the results in two
different scenarios, comparing them with algorithms such as LEACH, PEGASIS,
LEACH_EX, and also the latest algorithms EECPK-means, RaCH, and C3HA
showed that the proposed algorithm produced better results in terms of dead node
count and overall system energy depletion. In the proposed algorithm, on average,
more nodes were alive in the network, and the remaining energy was at least 17%
higher than LEACH and other algorithms. Based on the results of the simulation
of the proposed algorithm, the following future studies can be considered: (1) The
sleep status of the nodes can be evaluated based on the distance between the nodes.
(2) The simulations can easily be applied in real-world cases to measure actual per-
formance. (3) Examination of the performance of other clustering algorithms in
selecting CH nodes. (4) Investigation of dynamic networks, where the position of
the nodes is not static and is constantly changing (network nodes and mainly the
BS). (5) Investigation of the performance of evolutionary algorithms, especially the
GA, to find the optimal parameters for the clustering algorithm and CHs. (6) Nodes
Energy‑aware clustering method for cluster head selection… Page 21 of 23 2

Fig. 17  A hypothetical 3D environment: a ink drops spreading with radius = 1 and b spreading with
radius = 1

can be distributed in a 3-D space (x, y, and z) instead of the traditional 2D space (x
and y), for example, the quadrotors network. Figure 17 shows a 3D representation of
ink drops spreading with different radii (Spherical functions).

Author contributions "A.B. and C. wrote the main manuscript text. A. prepared figures. All authors
reviewed the manuscript".

Funding The author(s) received no financial support for the research, authorship, and/or publication of
this paper.

Data availability No datasets were generated or analyzed during the current study.

Declarations
Conflict of interest The authors declare no competing interests.

Ethical approval This manuscript has not been published nor is it currently under consideration for pub-
lication elsewhere.

References
1. Narayan V, Daniel A, Chaturvedi P (2023) E-FEERP: enhanced fuzzy based energy efficient routing
protocol for wireless sensor network. Wireless Pers Commun 110:1–28
2. Mohammed FAB et al (2022) Sectored LEACH (S-LEACH): an enhanced LEACH for wireless sen-
sor network. IET Wireless Sensor Syst 12(2):56–66
3. Yadav A, Kohli N (2021) Prolong stability period in node pairing protocol for wireless sensor net-
works. Int J Eng 34(12):2679–2687
4. Akyildiz IF et al (2002) A survey on sensor networks. IEEE Commun Mag 40(8):102–114
5. Shende MSS (2023) A review on wireless sensor network: its applications and challenges. Int J
Comput Res Eng Sci 1(01):18–25
 2 Page 22 of 23 E. Alimohammadi et al.

6. Vahabi S, Eslaminejad M, Dashti SE (2019) Integration of geographic and hierarchical rout-

ing protocols for energy saving in wireless sensor networks with mobile sink. Wireless Netw
25(5):2953–2961
7. Othman MF, Shazali K (2012) Wireless sensor network applications: a study in environment moni-
toring system. Proc Eng 41:1204–1210
8. Gopi P (2014) Multipath routing in wireless sensor networks: a survey and analysis. IOSR J Comput
Eng 16(4):27–34
9. Bahadur DJ, Lakshmanan L (2023) A novel method for optimizing energy consumption in wireless
sensor network using genetic algorithm. Microprocess Microsyst 96:104749
10. Dhouib S (2023) Hierarchical coverage repair policies optimization by Dhouib-Matrix-4
Metaheuristic for wireless sensor networks using mobile robot. Int J Eng 36(12):2153–2160
11. Yaro AS, Malý F, Malý K (2023) A two-nearest wireless access point-based fingerprint clustering
algorithm for improved indoor wireless localization. Emerg Sci J 7(5):1762–1770
12. Sheikhpour R, Jabbehdari S, Khadem-Zadeh A (2011) Comparison of energy efficient clustering
protocols in heterogeneous wireless sensor networks. Int J Adv Sci Technol 36:27–40
13. Harun, H.B., M.S. Islam, and M. Hanif (2022) Genetic algorithm for efficient cluster head selection
in LEACH protocol of wireless sensor network. In: 2022 International Conference on Advancement
in Electrical and Electronic Engineering (ICAEEE). IEEE
14. Qiu Y, Ma L, Priyadarshi R (2024) Deep learning challenges and prospects in wireless sensor net-
work deployment. Arch Comput Methods Eng 31(6):1–24
15. Purkar SV, Deshpande RS (2020) Clustering algorithm for deployment of independent heterogene-
ous wireless sensor network. Wireless Pers Commun 112(2):1303–1317
16. Sohail A (2023) Genetic algorithms in the fields of artificial intelligence and data sciences. Annals
Data Sci 10(4):1007–1018
17. Faris M et al (2023) Wireless sensor network security: a recent review based on state-of-the-art
works. Int J Eng Bus Manag 15:18479790231157220
18. Heinzelman, W.R., A. Chandrakasan, and H. Balakrishnan (2000) Energy-efficient communication
protocol for wireless microsensor networks. In: Proceedings of the 33rd annual Hawaii international
conference on system sciences. IEEE
19. Rajaram V et al (2024) Enriched energy optimized LEACH protocol for efficient data transmission
in wireless sensor network. Wireless Netw. https://​doi.​org/​10.​1007/​s11276-​024-​03802-5
20. Haghzad Klidbary S, Javadian M (2024) Improvement of low energy adaptive clustering hierarchi-
cal protocol based on genetic algorithm to increase network lifetime of wireless sensor network. Int
J Eng 37(9):1800–1811
21. Parwekar P, Rodda S (2017) Optimization of clustering in wireless sensor networks using genetic
algorithm. Int J Appl Metaheuristic Comput (IJAMC) 8(4):84–98
22. Ahmad R et al (2024) Optimization algorithms for wireless sensor networks node localization: an
overview. IEEE Access 12:50459–50488
23. Singh SK, Singh M, Singh D (2010) A survey of energy-efficient hierarchical cluster-based routing
in wireless sensor networks. Int J Adv Netw Appl (IJANA) 2(02):570–580
24. Shi, S., X. Liu, and X. Gu (2012) An energy-efficiency Optimized LEACH-C for wireless sensor
networks. In: 7th international conference on communications and networking in China. IEEE
25. Anand, G. and R. Balakrishnan (2013) Leach-Ex protocol-A comparative performance study and
analysis with leach variants of wireless sensor networks. In: IEEE Conference, Malaysia
26. Lindsey, S. and C.S. Raghavendra (2002) PEGASIS: Power-efficient gathering in sensor informa-
tion systems. In: Proceedings, IEEE aerospace conference. IEEE.
27. Linping, W., et al. (2010) Improved algorithm of PEGASIS protocol introducing double cluster
heads in wireless sensor network. In: 2010 International conference on computer, mechatronics,
control and electronic engineering. IEEE.
28. Jafri, M.R., et al., (2013) Maximizing the lifetime of multi-chain PEGASIS using sink mobility.
arXiv preprint arXiv:​1303.​4347
29. Ray A, De D (2016) Energy efficient clustering protocol based on K-means (EECPK-means)-mid-
point algorithm for enhanced network lifetime in wireless sensor network. IET Wireless Sensor Sys-
tems 6(6):181–191
30. Ngangbam R, Hossain A, Shukla A (2020) Improved low energy adaptive clustering hierarchy and
its optimum cluster head selection. Int J Electron 107(3):390–402
31. Kardi A, Zagrouba R (2020) Rach: a new radial cluster head selection algorithm for wireless sensor
networks. Wireless Pers Commun 113:2127–2140
Energy‑aware clustering method for cluster head selection… Page 23 of 23 2

32. Tay M, Senturk A (2022) A new energy-aware cluster head selection algorithm for wireless sensor
networks. Wireless Pers Commun 122(3):2235–2251
33. Jokar E et al (2020) Hardware-algorithm co-design of a compressed fuzzy active learning method.
IEEE Trans Circuits Syst I Regul Pap 67(12):4932–4945
34. Javadian M, Hejazi A, Klidbary SH (2022) Obtaining fuzzy membership function of clusters with
the memristor hardware implementation and on-chip learning. IEEE Trans Emerg Top Comput
Intell 6(4):1008–1025
35. Murakami M, Honda N (2007) A study on the modeling ability of the IDS method: a soft computing
technique using pattern-based information processing. Int J Approx Reason 45(3):470–487
36. Klidbary SH, Shouraki SB, Linares-Barranco B (2019) Digital hardware realization of a novel adap-
tive ink drop spread operator and its application in modeling and classification and on-chip training.
Int J Mach Learn Cybern 10:2541–2561
37. Klidbary, S.H., et al. (2017) Outlier robust fuzzy active learning method (ALM). In: 2017 7th inter-
national conference on computer and knowledge engineering (ICCKE). IEEE
38. Klidbary SH, Shouraki SB, Afrakoti IEP (2019) An adaptive efficient memristive ink drop spread
(IDS) computing system. Neural Comput Appl 31:7733–7754
39. Klidbary SH, Shouraki SB (2018) A novel adaptive learning algorithm for low-dimensional feature
space using memristor-crossbar implementation and on-chip training. Appl Intell 48(11):4174–4191

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under
a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted
manuscript version of this article is solely governed by the terms of such publishing agreement and
applicable law.