Improved Particles Swarm Optimization Based Clustering Process and Data Transmission for Energy Conservation in Wireless Sensor Networks

Recently, wireless communications have gained prominence due to their strong and versatile capabilities for information transmission. Wireless communications denote the relationship between mobility and connection, utilising the atmosphere as a transmission channel [1]. WSN are the predominant technology among current wireless systems. The operation of wireless sensor networks requires effective computational and energetical operations. In these circumstances, communication protocols serve as systematic frameworks that ensure the efficacy of these activities. A WSN comprises a collection of electromechanical devices spread throughout a specified geographical area. These components can furnish significant insights regarding the region of their distribution. This data is exchanged between network sensors via a transmission channel [2]. Battery management affects wireless sensor performance. Since it influences energy autonomy, this is crucial. Information processing is improved by BSs' centralisation. WSNs term the BS the sink or fusion centre for its aggregation.

Popular routing techniques with intriguing features like scalability and efficiency in communications are cluster-based or hierarchical routing. WSN energy conservation methods include hierarchical routing ideas. Hierarchical processing and transmission are only possible with high-energy nodes [3]. In contrast, low-energy nodes collect data near the aim. Hierarchical routing reduces cluster energy utilisation by aggregating information. Hierarchical routing consolidates processes to reduce sink transmission packet size [4]. Nodes may extend network lifespan and architectural scalability when given tasks. Specialised nodes serve as cluster-based or hierarchical reference points. Scalability matters in WSNs. Early assumptions prevent many routing algorithms from solving this challenge. Cluster-based protocols span the area with one sink and several cluster heads in a typical WSN design [5]. This design limits WSN scalability, resulting in expensive energy use. Increasing the network's nodes or impact breadth causes energy overload and information transmission bottlenecks. Node energy conservation is essential for a good protocol. The goal is to increase battery charging frequency [6]. Recharging is difficult or impossible, thus this is crucial. This system distributes power consumption across nodes utilising internal software processes [7]. Each node performs two tasks in hierarchical routing. These operating modes save energy. When this architecture is used at all levels, cluster head and general sensor transmission modes may be significant [8]. Conversely, optimisation tactics like metaheuristic procedures pertain to techniques capable of addressing complicated systems. These algorithms are based on biological or social events, which might be considered search strategies. The literature presents numerous metaheuristic techniques due to its wide application possibilities [9]. DE, Genetic Algorithms, Artificial Bee Colony (ABC), Gravitational Search Algorithm (GSA), and Grey Wolf Optimiser are metaheuristic systems. No continuity, differentiability, convexity, or beginning conditions are assumed for metaheuristics [10]. These qualities show the main advantages over other optimising methods. Although promising, these methods still create hurdles when used to high multimodal formulations. Consequently, the contributions of this study are as follows:

To develop a hybrid clustering model that leverages PSO to optimize the selection of medoids in K-Medoids, reducing sensitivity to initial medoid selection [11]. Especially, under adaptive K-Medoids-PSO model the number of clusters (K) is dynamically determined instead of being predefined.

To adopt priority-based TDMA scheduler that assigns slots based on QoS parameters such as delay, reliability, and packet size [12].

The subsequent portions of the paper are organised as follows. The related work is detailed in Section 2. Section 3 delineates our clustering algorithm. Section 4 presents the numerical simulations of this methodology in comparison to other established methods. The findings are addressed in Section 5.

The developed clustering approach uses two evolutionary algorithms. CH and cluster member nodes are selected using PSO. Euclidean distance determines the optimal CH sensor-sink route. Figure 1 demonstrates how the source node sends data to the sink node after finding the optimum CH-sink path.

1.png

Figure 1. Architecture of the proposed scheme

3.1 Network model

This research analyses the random placement of n sensor nodes in an M×M square area (A). Network premises are as follows.

• All SN share the same beginning energy.

• SN use GPS or other localization technologies to determine their position.

• All SN are fixed after deployment.

• All SN have the same transmission range and are aware of their leftover energy.

• SN modify energy usage depending on receiver distance.

• Each node has a distinct ID.

• The static BS is located at the square area's edge.

3.2 Energy model

Most radio nodes require energy at the transmitter, power amplifier, and receiver. The model employs transmitter-receiver distance-based free space and multi-path fading channels. Node energy consumption is proportional to d^2 if propagation distance d is less than threshold distance d_{_}0, else d^4. Eq. (1) calculates the energy required to convey a 1-bit packet d from transmitter to receiver,

${{E}_{T}}\left( l,d \right)=l.{{E}_{elec}}+\left\{ \begin{matrix}l\times {{\varepsilon }_{fs}}\times {{d}^{2}},d<{{d}_{0}}  \\l\times {{\varepsilon }_{mp}}\times {{d}^{4}},d\ge {{d}_{0}}  \\\end{matrix} \right.$     (1)

where, ${{E}_{elec}}$ is the energy used by an SN to send or receive 1-bit data, while ${{\varepsilon }_{fs}}$ and ${{\varepsilon }_{mp}}$ are the free-space and multi-path fading amplifier coefficients. Receive energy is given by

${{E}_{rx}}\left( l \right)=l.{{E}_{elec}}+{{E}_{wake\_rx}}$    (2)

Aggregation energy is given as

${{E}_{da}}\left( l \right)=l.{{E}_{da}}$    (3)

Eq. (2) calculates ${{d}_{0}}$,

${{d}_{0}}=\sqrt{{{\varepsilon }_{fs}}/{{\varepsilon }_{mp}}}$    (4)

Here, $\varepsilon_{fs}$ and $\varepsilon_{mp}$ are amplifier parameters. The free-space model is used for sensor node energy consumption when $\mathrm{d}<\mathrm{d}_0$, using $\varepsilon_{fs}$ as the amplifier value. Sensor node energy consumption is calculated using the multi-path fading model and amplifier value $\varepsilon_{mp}$ when $\mathrm{d} \geq \mathrm{d}_0$. This study limits node transmission distance to ${{d}_{0}}$ to guarantee that nodes in the same cluster are within the transmission range of the proposed PSO-based uneven dynamic clustering technique.

3.3 Proposed K-Medoids IPSO based clustering

Each sensor node cluster in the wireless sensor network receives a CH. The BS receives cluster node data from the CH. This work proposes K-Medoids WSN clustering. All sensors are clustered using K-Medoids. Clusters of K-Medoids represent one item. Medoid centres cluster. K-Medoids connected to the cluster node determine the middle and shortest distance between clusters. The sensor nodes communicate better, consume less energy, and detect cluster centres, reducing packet delays. K-Medoids are efficient and converge at step count. K-Medoids IPSO clustering follows these phases:

Step 1: Select k input data points at random.

Step 2: Each data point goes to the cluster with the nearest centre.

Step 3: Add all cluster ‘i’ data points' distances. Specify the i cluster centre point to reduce estimated distance from other locations.

Step 4: Repeat 1 and 3 until convergence, or the center point stops moving.

Each WSN grouping requires sensor node clustering and CH selection. The CH's major task is sending cluster node data to the BS. Finding precise centroids using K-Medoids reduces power consumption, packet latency, and sensor node performance. The problem is addressed by choosing the bestone from numerous solutions. In PSO, target value or condition, gbest, and stopping value are always recorded. Every PSO particle contains this information:

• Global solution data

• Velocity value indicates data change quantity

• Best value

3.3.1 Cluster formation

The cluster is established by the base station or sink via centralised clustering. The BS sends information gathering messages to all sensor nodes, which send node information to the BS, including node id, location (X and Y distance from the BS), energy loss and energy loss ratio (velocity), and current energy:

Step 1. Transformation of the issue into the PSO framework, whereby each PSO particle has two dimensions: location and velocity.

Step 2. Fitness function valuation. The proposed PSO-based clustering fitness function maximises member node average distance and energy relative to the cluster head and headcount. The formula below calculates particle fitness:

$fitness~value=Fv={{\propto }_{I}}$    (5)

$=\frac{\mathop{\sum }_{i=0}^{n}d\left( current~node,member~i \right)}{n}+{{\propto }_{2}}$    (6)

$=\frac{\mathop{\sum }_{i=0}^{n}E\left( member~i \right)}{E\left( current~node \right)}$$+\left( 1-{{\propto }_{1}}-{{\propto }_{2}} \right).\frac{1}{\begin{matrix}number~of~members~  \\covered~by~current~node  \\\end{matrix}}$    (7)

where, $\propto_1$ and $\propto_2$ represent the weighting parameters and n indicates the number of individuals included inside the cluster.

Step 3. New particle formation from solution. Creating new particles from existing ones is novel particle generation.

A particle's current velocity is used to estimate its new velocity, which is the rate of change in its position. New velocity is calculated as follows:

$new~velocity={{\omega }^{*}}old~velocity+{{\omega }_{1}}\left( local~best~position-current~best~position \right)+{{\omega }_{2}}\left( global~best~position-current~best~position \right)$    (8)

Here, ω represents the inertia weight, whereas ${{\omega }_{1}}$ and ${{\omega }_{2}}$ denote fundamental tuning parameters of PSO. Inertia weight ω ontrols how much of the previous velocity is retained. A higher value encourages global exploration, while a lower value promotes local exploitation. Cognitive coeffecint ${{\omega }_{1}}$ affects the particle's inclination to return to its favourite spot. Social coeffecient ${{\omega }_{2}}$ guides the particle toward the global best-known position. A commonly used setting includes a constriction factor or values such as ω=0.729, ${{\omega }_{1}}={{\omega }_{2}}=1.49445$ ensure convergence stability.

The estimation of the particle's new location is as follows:

$new~position=old~position+new~velocity$    (9)

Step 4. The fitness function in Step 2 calculates the fitness value for new particles using their velocity and position.

Step 5. The following iteration uses the best particle fitness value from the old and new ones.

New fitness value > previous fitness value

select new particle

else

old particle is forwarded to next iteration

Step 6. There is one local best answer every iteration. The particle having the highest fitness in the current iteration is the local optimal solution.

Step 7. The particle's global best solution is its maximum local best solution throughout all repetitions. Solutions cluster. SN receives a base station cluster-announcement message that creates the cluster via PSO. Every node begins a CH selection cycle after receiving this message. Each CM chooses a CH to form efficient and balanced clusters based on a unique cost function that considers distance to CH, residual energy, and CH member deviation, illustrated as:

$cos{{t}_{t}}=\frac{Eres\left( i \right)}{{{D}_{t0}}CH\left( j \right)\times \left| ND\left( C{{H}_{j}} \right)-\frac{\mathop{\sum }_{j=1}^{k}ND\left( C{{H}_{j}} \right)}{k} \right|}$    (10)

where, D_t0 CH(j) represents the distance among CM(i) and its candidate (CH)_j, and k represents the number of CHs. Additionally, ND(CH_j) represents CM(i)'s candidate's node degree. First, the final picked CH broadcasts its ID and node degree inside its communication range. The node degree is unitless, but it is being multiplied by a cost term with energy units (e.g., Joules/bit). Each CM estimates cost. The CM will proclaim itself as a CH when k=0, meaning no CH message is received. Therefore, optimum clusters are constructed with higher-cost CHs and CMs. Data transfer after cluster formation is particularly important for WSN energy saving since inappropriate transmission causes failure or collision. Transmission from CMs to CHs using time division multiple access (TDMA) improves these issues.

Algorithm 1: Clustering using K-Medoids with a Particle swarm optimization

Network Initialization

Step 1: Initialize of the WSN S={s1,s2,…sn}

Step 2: Place BS at (50, 180)

Step 3: Place all the SNs

K-Medoids clustering and IPSO CH selection, initiate initial K range: K $\in$ [2, 5]

S: Number of swarms; 4 particles

P: Number of particles in each swarm;

X × Y: Dimensions of network terrain

Maximum iteratens=5

Position X_i and velocity V_i for each particle (node) i

Randomly initialize k- medoids (candidate cluster heads) for each particle

Calculate the euclidian distance and compute fitness function for each particle

fitness=∑_(i=1)^k ∑_(s_j ϵC_i) d(s_j,mediod (i)

Update velocity for each particle and find new position

     Convergence criteria: No significant improvement in fitness over $T$ iterations, Minimal shift in medoid locations.

     Termination based on stability: Stability of $k$ over iterations

If stopping criteria are met then terminate. Otherwise, go back to Step 2

Finalize optimal medoids as cluster heads and start data transmission by schedulling

3.4 Data transmission process by scheduling

When many CH nodes carry a packet of data, it will collide, and forwarding to a less-than-ideal CH node diminishes WSN efficiency. We must give each CH node a forwarding priority and ensure that the best one may send packets first to enable packets to go via the worldwide optimum routing channel and avoid packet collisions. A greater U value results in a shorter holding time in this technique. For temporal efficiency, we employ data transfers between neighbouring nodes, consider node i's relative U value, and create as:

$f\left( U \right)=\left\{ \begin{matrix}U-{{U}_{last}},U<{{U}_{last}}~such~that~U<0  \\0,U\ge {{U}_{last}}  \\\end{matrix} \right.$    (11)

where, U_last is the ideal relay node sender's last transmission U value. The minimum propagation delay (t_(max)) with propagation speed v is

$t\left( max \right)=\frac{R}{v}$    (12)

The waiting time ensures that node 2 receives the packet from node 1 before passing it and reduces duplicated transmission.

$\delta \ge tanh\left[ -f\left( {{U}_{1}} \right) \right]-tanh\left[ -f\left( {{U}_{2}} \right) \right]$    (13)

The retention duration of each node i is determined by

${{t}_{i}}=tanh\left[ -f\left( {{Q}_{i}} \right) \right].\left( \frac{2{{t}_{max}}}{\delta } \right)$    (14)

Initially, devices with 0 or less energy were considered dead and unable to deliver data. If they are in the base station's transmission range and have the highest U-value in their region, devices may connect directly without an intermediate device. If the CH is likewise far away, devices remote from the base station may nevertheless submit packets to it via a neighboring cluster device.