An innovative cluster-based power-aware protocol for Internet of Things sensors utilizing mobile sink and particle swarm optimization

Abstract

Over the past decade, the Internet of Things (IoT) has become a necessary technology increasingly applied in many fields of research and development, such as smart cities and homes, health, industry, agriculture, security, and surveillance. In IoT systems, the use of sensors is considered an important common manner in which all devices communicate with a wireless sensor network to form an information system that comprises a massive number of sensor nodes performing accurately to create a smart decision-making method. However, these sensor nodes might be employed in severe environments, where replacing or recharging their batteries is considered an impossible mission. Simultaneously, the limitation of energy resources in sensor nodes presents a challenging issue that reduces the lifespan of individual nodes and the overall network system as a result of energy depletion. These obstacles necessitate energy-efficient routing protocols. According to the literature review, various routing protocols have been introduced, especially those that use clustering techniques. However, they have many drawbacks due to the way of selecting the cluster head (CH), which results in consuming energy dramatically, and consequently shortening the network lifetime. Additionally, instead of using the static sink, which was inefficient in collecting data, many researchers studied the behavior of the mobile sink (MS), which also has many downsides that negatively impact network performance. This paper presents a novel energy-conscious protocol for clusters that incorporates an adaptive movement for mobile stations and utilizes particle swarm optimization (PSO). The circular network area is divided into clusters, each of which has an elected CH based on the PSO technique. The MS aims to distribute energy among nodes to prevent hotspot issues. To achieve better coverage, it moves in a circular pattern with a constant angular velocity, starting from the center of the network area and moving forward and backward along the radius of the network. This research conducts intensive simulations, which run on MATLAB R2018, to assess the performance of our proposed protocol and compare its results with those of pertinent works. The results obtained are encouraging and demonstrate that the protocol we proposed surpasses its counterparts by significantly extending the lifespan of the network.

Appendices

Appendix A: CH selection using PSO example

In the initialization process of PSO, the nodes in each cluster get arranged in a matrix that contains the ID, the initial position, the initial velocity, and the energy of the nodes. Every particle announces itself as a $P_{best}$, and then it searches for the $G_{best}$ solution by calculating the fitness function $(F)$, as mentioned previously in Eqs. 5–7. For example, in cluster 1, as shown in Fig. 

Fig. 11

Nodes distribution for cluster 1

11, there are groups of 20 nodes. Therefore, we use the predefined swarm size of 20 particles, and the MS is situated in the position of (25, 25) (Table

Table 4 Equations’ parameters

4). Moreover, Table

Table 5 Initial cluster nodes’ details

5 shows all nodes’ details, including their ID, position, residual energy, initial velocity, initial $P_{best}$, and calculated fitness based on Eqs. 5–7. However, Table 4 represents the main parameters utilized in the PSO equations, bearing in mind that these parameters have been selected after conducting a massive number of simulations and showed a noticeable and remarkable network performance enhancement.

For clarity, suppose we compute the fitness function $(F)$ of the first node, whose position is (5.1, 9.36), then $f_{1}$ is 0.747, and $f_{2}$ is 0.52, referring to Eqs. 6 and 7, respectively, detailed in Sect. 4.5. Accordingly, and based on Eq. 5, $F_{1}^{1} = 0.633$. The fitness of the remaining nodes is determined in a way that resembles the calculation of the first node’s fitness, as shown in Table 5. Afterward, the $G_{best}$ solution is found based on the maximum fitness value (i.e., 0.639) between all nodes, in that case, node 13 is considered the initial $G_{best}$(i.e.,$G_{{best_{x} }}^{1} = 7.6865,G_{{best_{y} }}^{1} = 16.48$).

As soon as the initialization phase ends, the PSO forms a fitness matrix of all nodes in the cluster at each iteration. Noting that, the particles are initialized by assigning coordinates to the nodes. The fitness matrix consists of nineteen columns as follows: F(col. 1) is the ID of the node, F(col. 2) and F(col. 3) are the current positions of a node on the x-axis ($P_{x}^{i}$) and y-axis ($P_{y}^{i}$), in sequence. F(col. 4) indicates the ratio of the residual energy of a node to its initial energy. However, the current velocity of a node on the x-axis ($V_{x}^{i}$) and y-axis ($V_{y}^{i}$) is represented in F(col. 5) and F(col. 6), in a row. The $P_{{best_{x} }}^{i}$ and $P_{{best_{y} }}^{i}$ are the location of the current $P_{best}^{{}}$ solution of a node, and they are shown in F(col. 7) and F(col. 8). Interestingly, F(col. 9) is the fitness of the $P_{best}^{{}}$ solution proposed by the ID^th node. F(col. 10) and F(col. 11) represent two generated random variables $(r_{1} ,r_{2} )$. F(col. 12) and F(col. 13), which can be computed referring to Eqs. A.1 and A.2, are the updated velocities for the next iteration on x-axis($V_{x}^{i + 1}$) and y-axis( $V_{y}^{i + 1}$). However, the updated positions for the next iteration on the x-axis ($P_{x}^{i + 1}$) and y-axis ($P_{y}^{i + 1}$) are expressed in F(col. 14) and F(col. 15), which can be computed referring to Eqs. A.3 and A.4. F(col. 16) is the computed fineness after update the positions of a node. In addition to that, F(col. 17) and F(col. 18) represent the updated $P_{best}^{{}}$ solution after the comparison is made. In other words, if the $F^{i} (P_{best} ) \ge F^{i + 1} (P_{best} ),$ then the $F^{i} (P_{best} )$ will remain the same without change, otherwise, $F^{i} (P_{best} )$ is updated to become the value of $F^{i + 1} (P_{best} )$, which is clearly denoted in F(col. 19), and then the algorithm searches now for the $G_{best}$. It is worth mentioning that, in general, to find the $G_{best}$ solution (i.e., the maximum value of F(col. 19)), during each iteration, every particle uses its own $P_{best}$ and $G_{best}$ solution to update its position and velocity in order to reach the $G_{best}$, which are detailed in Eqs. A.1–A.4, shown below:

$$F(col. 12) = \omega \times F(col. 5) + c_{1} \times F(col. 10) \times (F\left( {col.7} \right) - F\left( {col. 2} \right)) + c_{2} \times F(col.11) \times (G_{{best_{x} (t)}} - F(col. 2))$$

(A.1)

$$F(col. 13) = \omega \times F(col. 6) + c_{1} \times F(col. 10) \times (F\left( {col.8} \right) - F\left( {col. 3} \right)) + c_{2} \times F(col.11) \times (G_{{best_{y} (t)}} - F(col. 3))$$

(A.2)

$$F(col. 14) = F(col. 2) + F(col. 12)$$

(A.3)

$$F(col. 15) = F(col. 3) + F(col. 13)$$

(A.4)

• Iteration 1

To this point, we will start iteration 1, where the concept of particles arises. Every particle considers itself as the best personal solution and then it searches for the $G_{best}$ solution. It is noteworthy to mention that a new velocity and position for each particle is found for each iteration based on Eqs. A.1–A.4. As a result, Table

Table 6 The current and updated values of PSO at iteration 1

6 shows the fitness matrix at iteration 1. Referring to the entries of Table 6, every node updates its velocities and positions based on the previous velocities and positions. After that, the new fitness is computed. If the new fitness is lower than (or the same as) of the fitness of the $P_{best}$ solution (i.e.,$F^{1} (P_{best} )$, then there will be no change in this solution. Particularly, the $P_{best}$ will remain as is (like the nodes 13 and 14). On the other hand, if the new computed fitness of a node is greater than $F^{1} (P_{best} )$, then there will certainly be an update on that solution (like all nodes excluding nodes 13 and 14). At the end of each iteration, the algorithm compares the $F(G_{best} )$ with those of the updated fitness matrix solutions $(F^{2} (P_{best} ))$ . In other words, the $F(G_{best} )$ will be the node that has the highest fitness among the nodes. Accordingly, the $G_{best}$ solution in iteration 1 is updated to become node 19 based on the maximum fitness value (i.e., 0.661) between all nodes. Thus, $G_{best}$ (i.e.,$G_{{best_{x} }}^{2} = 36.347,G_{{best_{y} }}^{2} = 22.090$).

• Iteration 2

As far as the second iteration is concerned, a new velocity and position for each particle are found based on Eqs. A.1–A.4. Therefore, Table

Table 7 The current and updated values of PSO at iteration 2

7 summarizes the fitness matrix at iteration 2.

To this point, the matrix of 20 nodes for the PSO algorithm is filled with the current positions, velocities, $P_{best}$ on the x-axis and y-axis, as well as the $F^{2} (P_{best} )$. It must be pointed out that these values are the ones that were generated in the prior iteration. After the two random numbers are generated, the new positions and velocities are evaluated. Based on these updated values, the fitness function $(F)$ is estimated. In view of that, the comparison is made between this fitness function $(F_{i}^{3} )$ and the $F_{i}^{2} \left( {P_{best} } \right)$ for each particle. In different words, if $F_{i}^{3}$ is higher than $F_{i}^{2} \left( {P_{best} } \right)$, then the $P_{best}$ will be updated based on the updated positions (i.e., $P_{{best_{x} }}^{3} = P_{x}^{3}$ and $P_{{best_{y} }}^{3} = P_{y}^{3}$). On the other hand, the $F\left( {P_{best} } \right)$ will stay as it, if $F_{i}^{3}$ is higher than $F_{i}^{2} \left( {G_{best} } \right)$.

The $G_{best}$ also will be checked out at the end of this iteration. It is known that it was updated during the prior iteration, which becomes referring to node 19, which has a fitness of 0.661. At the end of the second iteration, the greatest fitness is 0.685, which belongs to node 1. Therefore, $G_{best}$ is updated since this node has the maximum fitness value, which has new positions equal to $G_{{best_{x} }}^{3} = 14.632,G_{{best_{y} }}^{3} = 14.154.$.

In this fashion, the PSO algorithm continues the process of searching for the $G_{best}$ solution. To keep track of the remaining iterations, we provide the readers with a set of iterations. For instance, Tables

Table 8 The current and updated values of PSO at iteration 3

8,

Table 9 The current and updated values of PSO at iteration 4

9,

Table 10 The current and updated values of PSO at iteration 5

10,

Table 11 The current and updated values of PSO at iteration 6

11,

Table 12 The current and updated values of PSO at iteration 7

12,

Table 13 The current and updated values of PSO at iteration 8

13,

Table 14 The current and updated values of PSO at iteration 98

14,

Table 15 The current and updated values of PSO at iteration 99

15 and

Table 16 The current and updated values of PSO at iteration 100

16 show the fitness matrix for iterations 3, 4, 5, 6, 7, 8, 98, 99, and 100. For the last iteration, node 1 is the $G_{best}$, which is the one that has the highest fitness (i.e., 0.696) among all nodes. Thus, for cluster 1, node 1 becomes the new CH, and all data from other nodes are forwarded to it.

Appendix B: illustrative example of MS circular movement

In this appendix, we provide an illustrative example that demonstrates the MS movement. First, suppose that the number of clusters ($C_{n}$) in the network is equal to 5, then the cluster angle $\left( A \right)$ is 1.26 radians, based on Eq. 1. In addition, $R_{N}$ is equal to 400. Then, depending on the RT, $R_{ms}$ is variant from $[100,200,300]$ in the forward direction and [300, 200, 100] in the reverse direction. Also, assume the constant angular velocity is equal to $\pi /20$ radians/sec, initial position ($P_{{ms_{{_{0} }} }}$) is (0,0). To begin, we can make an initial assumption that the ratio in Eq. (2) is equal to 1. This means that the result of Eq. (2) will be 45° or $\pi /4$. We can then use Eqs. (2), (3), and (4), as shown in Table

Table 17 The positions of MS node based on the RT and angle change

17, to obtain the circular motion of the MS node, as depicted in Fig. 

Fig. 12

The circular motion of MS at $R_{ms}$ = 100 and the change of RT on the interval [0,400]

12. For this particular example, we assume that RT is a value between 0 and 400, with $R_{ms}$ set at 100. As mentioned previously, $R_{ms}$ is changed every 400 rounds. It is worth noting that if our protocol runs in the next interval between 401 and 800, then $R_{ms}$ will be equal to 200, and we can continue our calculations to obtain the circular motion of the MS with varying $R_{ms}$ and RT intervals.