Energy-efficient data gathering protocol in unequal clustered WSN utilizing fuzzy multiple criteria decision making

Abstract

Due to the imbalanced energy consumption among nodes, some nodes die prematurely, which decreases the network lifetime in wireless sensor networks, especially many nodes and one Sink. To solve this problem, in this paper, we propose an energy-efficient data gathering protocol in unequal clustered WSN utilizing fuzzy multiple criteria decision making called DGUCF. In DGUCF, nodes can be divided into unequal clusters to balance energy consumption in the network scenario with nodes of non-uniform distribution and heterogeneous energy. First, DGUCF adopts an intuitionistic fuzzy analytic process and hierarchical fuzzy integral to select cluster heads synthetically. Then, we propose a new cluster head competition radius that is fit for this network scenario. In succession, to balance energy consumption, each cluster head transmits data packets over the next hop cluster heads in proportion to their residual energy. Finally, the simulation results demonstrate that the DGUCF protocol can balance energy consumption effectively, and the network lifetime can be prolonged.

Keywords

Wireless sensor networks unequal cluster cluster head intuitionistic fuzzy analytic hierarchy process fuzzy multiple criteria decision making

1 Introduction

Wireless Sensor Networks (WSNs) have been widely used in civil and military fields, such as smart home, bio-medical, environmental monitoring, machinery manufacturing, space exploration, etc. [1]. With sensor node integration and miniaturization, most recent sensor nodes have energy limited batteries. Since nodes are usually located in unmanned areas or hazardous environments, it is difficult to supply energy. Renewable energy technology is not yet mature. Therefore, it is the primary challenge to maximizing the network life of a wireless sensor network by optimizing the system energy consumption [2].

Recently, the clustering data gathering protocol is one of the effective methods to reduce energy consumption and prolong the network lifetime [3]. In the clustering gathering protocol, a cluster head (CH) must be selected as a manager of the cluster, others are the member nodes of the cluster. Because the cluster head is responsible for receiving, compressing and forwarding data from its member nodes, its energy consumption is higher than that of the cluster members. To resolve this problem, most clustering protocols divide time periods into rounds and periodically rotate nodes as cluster heads to balance the energy consumption among the nodes. However, despite the periodic rotation of the cluster heads, the imbalance of energy consumption due to the different distances between the cluster heads is another problem.

When a cluster head is far from the BS that transmits data to it, the distance from the node to the BS is too long for a single hop network. Hence, the energy consumption of these cluster heads is higher than that of cluster heads near the BS. In a multi-hop network, a cluster head near the BS is responsible for relaying an amount of data that results in higher energy consumption. This imbalance of energy consumption can lead to the early death of a node and result in a network separation. To resolve this problem, researchers use an unequal clustering protocol to balance the energy consumption in WSNs.

Thus far, researchers have provided several unequal clustering data gathering protocols in WSNs. However, most of them assume that nodes are uniformly deployed and have the same energy. There are few unequal clustering data gathering protocols that consider both the random non-uniform distribution of nodes and the heterogeneous energy of nodes [4].

Non-uniform clustering size protocols have been studied extensively recently. An unequal clustering size (UCS) protocol [5] was the first to adopt an unequal cluster size to gather data. In a network model comprised of rings, UCS acquires the expected number of clusters in different rings through which the cluster competition radius is obtained. However, UCS is appropriate for heterogeneous WSNs, although, due to the use of a network model with two rings, UCS cannot effectively gather data in a large-scale WSN. Chen et al. [6] proposed the unequal cluster-based routing protocol. This protocol is a classical unequal clustering data gathering protocol. The clusters near the sink have a smaller cluster size, whereas the clusters far away from the sink have a larger cluster size. This ensures that cluster heads near the sink have sufficient energy to relay data from other CHs, which balances the energy consumption of the entire network. Lai et al. [7] explored the arranging cluster sizes and transmission ranges for wireless sensor networks (ACT). In ACT, the network is divided into many layers with different widths to prolong the network lifetime. The aim is to reduce the size of the clusters near the BS because CHs closer to the BS need to relay more data. The proposed method allows every CH to consume approximately the same amount of energy so that the CHs near the BS do not exhaust their power so quickly. However, ATC clusters according to the global information that affects the scalability of the clustering protocol. ACT is fit for WSNs with nodes of uniform deployment and the same energy. Raju Dutta et al. [8] proposed a new low-energy adaptive unequal clustering protocol using Fuzzy c-means in wireless sensor networks (LAUCF), which is an unequal clustering size model for the organization of networks based on a Fuzzy c-means (FCM) clustering algorithm that can lead to a more uniform energy dissipation among the cluster head nodes, thus increasing the network lifetime. Although LAUCF can prolong the network lifetime to a certain extent, the algorithm complexity is high.

During cluster head selection, whether a node can become a cluster head not only considers its residual energy but also other factors, such as the energy consumption, the number of neighbour nodes, etc. Therefore, a cluster head selection problem can be abstracted into a multi-criteria decision problem. Gao et al. [9] proposed a distributed energy-efficient clustering algorithm based on a fuzzy multiple criteria decision making approach [10]. In the process of cluster-head selection, FAHP considers the influence of many factors and avoids using an OWA operator to make each attribute independent of each other.

There are some limitations when FAHP employs the trapezoidal fuzzy analytic hierarchy to determine the importance of the criterion. A fuzzy set in the fuzzy analytic hierarchy process can express the subjective judgement of the decision-maker to a certain degree but cannot express the circumstance of abstention or indecision [11]. The Intuitionistic Fuzzy Analytic Hierarchy Process (IFAHP) is an extension and development of the fuzzy analytic hierarchy process (FAHP). IFAHP is more flexible and practical in addressing the above-mentioned uncertainties. Bagci et al. [12] explored a fuzzy energy-aware unequal clustering algorithm (EAUCF). EAUCF aims to decrease the intra-cluster work of the cluster-heads that are either close to the BS or have low remaining battery power. A fuzzy logic approach is adopted to handle uncertainties in cluster-head radius estimation. Although EAUCF adopts a fuzzy logic system to obtain the competition radius of cluster heads, it is only appropriate for scenarios with homogeneous nodes and independent decision attributes. Mao et al. [13] proposed an improved fuzzy unequal clustering scheme for large scale wireless sensor networks (IFUC). IFUC uses a fuzzy logic system to determine the probability of a cluster head and estimate the competition radius of the cluster head; then, the network nodes are divided into unequal clusters. Ant colony optimization (ACO) is adopted to construct energy-aware routing between the cluster heads and the base station. ACO reduces and balances the energy consumption of cluster heads and, to a large extent, solves the hot spots problem that occurs in multi-hop WSN routing protocols. However, IFUC only fits scenarios with homogeneous nodes and independent decision attributes. Hematkhah et al. [14] proposed a distributed clustering protocol called DCPVP, which is based on voting and priority ideas. In DCPVP, the size of the clusters is based on the distance of the nodes from the data link, such as the BS and local node density. The cluster heads are elected based on the mean distance from neighbours, remaining energy and the times elected as cluster head. Simulation results confirm that DCPVP improves the lifetime of networks and decreases the message complexity of the network. However, DCPVP only fits scenarios with homogeneous nodes, uniform node deployment and independent decision attributes.

Accordingly, the motivation of the paper is trying to develop a distributed energy-efficient data gathering algorithm in the network scenario with nodes of non-uniform distribution and heterogeneous energy. The paper makes the following main contributions:

We propose an energy-efficient CH selection scheme based on fuzzy multiple criteria decision making to improve the performance of the entire network.

During establishing unequal cluster topology, a new cluster head competition radius is explored in the case of heterogeneous energy and nodes of non-uniform distribution.

When data packets are transmitted between cluster heads, we devise a simple method to relay the aggregated data in proportion to two factors, which are the distance between CHs and the residual energy. In addition to this, the proposed method tries to balance the relaying load of the CHs in order to balance their energy consumption.

2 Related work

By now, some work has been finished on improving energy efficiency based on the uniform clustering protocols. However, there are only a few unequal clustering protocols improving energy with fuzzy logic system. Some clustering protocols are introduced in above section. In this section, we only briefly review some other typical clustering protocols and others adopting fuzzy logic system.

LEACH [15] was one of the most popular clustering algorithms for WSNs. In LEACH, each node independently elects itself as a cluster head with a probability. Heinzelman et al. [16] computed an optimal number of clusters to minimize total energy consumption of the whole network.

An energy efficient hierarchical clustering (EEHC) algorithm [17] was another clustering protocol, which divides the network into a hierarchy of layers. The hybrid energy-efficient distributed (HEED) protocol [18] set the election probability according to communication cost, neighbor density, or residual energy. Kumar et al. [19] proposed an energy efficient heterogeneous clustered (EEHC) scheme for WSNs based on weighted election probabilities of each node to become a cluster head according to the residual energy in each node. There are many clustering algorithm based on fuzzy mathematics, such as fuzzy logic system, fuzzy graph and fuzzy petri-nets.

Logambigai et al. [20] provided propose a fuzzy logic based unequal clustering algorithm in which final cluster is elected considering the energy level of the tentative cluster head within the cluster radius computed based on residual energy and node degree of the tentative cluster heads. Padmalaya et al. [21] explored an energy efficient clustering algorithm for WSNs using fuzzy logic concept. Seyyit et al. [22] proposed a multi-objective fuzzy clustering algorithm (MOFCA) which is not only energy-efficient but also distribution-independent for WSNs. However, the node distribution of the clustering algorithm is randomly uniform.

Jain et al. [23] proposed a new method for modeling and k - hop clustering for WSNs, which results in better scalability, balancing of energy consumption of nodes and longer network lifetime. Yu et al. [24] explored a reliable energy-efficient multi-level routing algorithm in Wireless Sensor Networks. The proposed algorithm considers the residual energy, number of the neighbors and centrality of each node for cluster formation, which is critical for well-balanced energy dissipation of the network. However, this cluster algorithm also establishes the equal cluster topology.

3 DGUCF protocol

In this section, we propose an energy-efficient data gathering protocol in unequal clustered WSN utilizing fuzzy multiple criteria decision making. Firstly, intuitionistic fuzzy AHP and hierarchical fuzzy integral are adopted to optimize the selection of cluster heads. In succession, we introduce the establishment of the unequal cluster topology. Finally, we present a new method for data transmission between cluster heads.

It is worthy to note that the network model in this paper is the same with [25]. Meanwhile, Atypical energy consumption model is adopted, the details of which can be found elsewhere [26].

DUCPMC functions according to a period of time denoted as a round, where each round comprises the following two phases: unequal clustering and data transmission. In the clustering phase, DUCPMC adopts a multiple criteria decision to select the cluster heads so that the network forms a cluster topology. In the data transmission phase, we propose an equal ratio routing algorithm to balance the energy consumption of the network. To save energy, the time of data transmission is longer than that of the cluster topology establishment.

3.1 Cluster head selection

In this subsection, intuitionistic fuzzy AHP and hierarchical fuzzy integral are adopted to optimize the selection of cluster heads. Firstly, cluster head evaluation hierarchy structure is designed. Furthermore, we introduce the attribute evaluation value of cluster head. Moreover, we determine the degree of criteria importance.

Energy status, QoS impact and location are taken into account simultaneously as the main factors that can influence the selection of cluster heads while each factor contains some sub-criteria. Fuzzy multiple attribute decision making is adopted to select optimal cluster heads by taking all factors into account synthetically. On the basis of the comprehensive analysis and the evaluation indices of the CH selection, the hierarchy structure is constructed as in Fig. 1.

Fig.1

Cluster head evaluation hierarchy structure.

3.1.1 The attribute evaluation value of cluster head (h)

In the initialization phase, every node broadcasts an initial message INITIAL _ MASSAGE to the network. This massage includes the node ID, residual energy and communication radius. The message is stored in the memory by the node receiving the message. The distance between the nodes is calculated according to the strength of the received message. When the distance between two nodes is smaller than the communication radius, they are considered neighbour node for each other and establish the neighbour table. After the initialization phase, every node calculates the maximal and minimal residual energy of its neighbour node, the number of neighbour nodes and the maximal and minimal distance to the BS. Therefore, every node obtains the expected parameters. In addition, every node calculates the attribute evaluation value (h).

Residual Energy (E)

In the self-organizing mechanism, the cluster-head consumes more energy than the cluster member nodes because the cluster head is primarily involved in data fusion, processing and routing. In WSNs, energy is the most important and scarce resource. We should first consider it. The evaluation value of residual energy is defined as follows. $E_{i} = \frac{E_{r} - E_{min}}{E_{max} - E_{min}}$ (1)

Here, E_max and E_min are the maximal and minimal residual energy, respectively, in the neighbour node of node i; E_r is the residual energy of node i. If the value of E_i is higher, the energy-criticality of the node is less.

Message Cost (M)

The message cost can be calculated by the number of messages from the initialization phase to the cluster topology establishment. The message cost directly effects the energy consumption of the nodes. Hence, the evaluation value of the message cost is defined as follows. $M_{i} = \frac{M_{T} - M_{C}}{M_{T}}$ (2)

Here, M_C is the total number of receiving and sending messages of node i; M_T is the total number of receiving and sending messages of a neighbour node. If the value of M_C is higher, the message cost lower.

Coverage Factor (CF)

Coverage maintenance is one of the most basic issues in guaranteeing quality of service (QoS). The coverage time of the network can be effectively prolonged by selecting nodes with better coverage as the cluster heads [30] and providing the nodes with more residual energy. Therefore, the coverage factor is the main factor for evaluating service quality. The evaluation value of the coverage factor is defined as follows. $M_{i} = \frac{M_{T} - M_{C}}{M_{T}}$ (3)

Here, |C (i) | denotes the number of neighbour nodes covered by node i; |O (i) | represents the set of neighbour nodes that are covered by node i and its neighbours. $| \cup_{i = 1}^{n} C (i) |$ is the total number of nodes covered by node i and all neighbour nodes of node i. If the value of CF_i is higher more nodes are covered only by node i and the coverage preservation of node i is better.

Link Reliability (R)

During data transmission, any node failure will result in many data packet losses. It is necessary for service quality to consider reliable data delivery. The cluster head is responsible for receiving data from the member nodes and other cluster heads. Hence, nodes with a high probability of high reliability should be cluster heads. The evaluation value of the link reliability is defined as follows. $R_{i} = \frac{B_{ava} (i)}{B_{total} (i)}$ (4)

Here, B_ava (i) is the available buffer space of node i; B_total (i) is the total buffer space of node i. If the available buffer space is larger, the reliability of node i is higher.

The number of neighbour node (N)

According to the initial information of the nodes, the optimal number of cluster heads can be obtained when the energy consumption of the nodes deployed uniformly is minimal. Hence, we can obtain the optimal number of cluster member nodes. Theoretically, the probability of becoming the cluster head is larger when the number of neighbour nodes is close to that of the optimal cluster member node. The evaluation value of the number of neighbour nodes is defined as follows. $N = \frac{| N_{i} - N_{o} |}{N_{o}}$ (5)

Here, N_i is the number of neighbour nodes for node i; N_o is the optimal number of cluster member nodes. It can be calculated by [16]. When the number of neighbour nodes equals the optimal number of cluster member nodes, the value of N equals one.

The distance to BS (D)

The distance between the node and BS is a main factor considered for the cluster head since distance and energy consumption have a direct relationship. If the distance between a cluster head and BS is longer, the probability of being a cluster head is less. Hence, the evaluation value of the distance to BS is defined as follows. $D_{i} = \frac{d_{max} - d_{(i, Sink)}}{d_{max} - d_{min}}$ (6)

Here, d_max is the maximal distance between the node and BS; d_min is the minimal distance between the node and BS; d_(i, Sink) is the distance between node i and BS. If the distance between node i and BS is longer, the value of D_i is lower.

3.1.2 Determining the degree of criteria importance (g)

The degree of criteria importance is determined by the intuitionistic fuzzy analytic process proposed in [31] and the cluster head evaluation hierarchy structure shown in Fig. 1. First, invited experts compare the importance of each criterion of the target layer with its first upper layer. Based on the opinions of the experts, the fuzzy complementary judgement matrix of intuition is established. Therefore, the intuitionistic fuzzy complementary judgement matrix for all attributes of the second layer to the target layer is as follows. $L_{1} - L_{2} = [\begin{matrix} 0.5 & (0.8, 0.9) & (0.7, 0.8) \\ (0.1, 0.2) & 0.5 & (0.4, 0.6) \\ (0.2, 0.3) & (0.4, 0.6) & 0.5 \end{matrix}]$ (7)

The intuitionistic fuzzy complementary judgement matrix for all attributes of the third sub-criteria layer to the second layer is as follows. $\begin{matrix} L_{21} - L_{3} \\ = [\begin{matrix} 0.5 & (0.8, 0.9) & (0.5, 0.7) & (0.6, 0.7) \\ (0.1, 0.2) & 0.5 & (0.1, 0.2) & (0.2, 0.3) \\ (0.3, 0.5) & (0.8, 0.9) & 0.5 & (0.5, 0.8) \\ (0.3, 0.4) & (0.7, 0.8) & (0.2, 0.5) & 0.5 \end{matrix}] \end{matrix}$ (8) $\begin{matrix} L_{22} - L_{3} \\ = [\begin{matrix} 0.5 & (0.4, 0.6) & (0.4, 0.7) \\ (0.4, 0.6) & 0.5 & (0.5, 0.7) \\ (0.3, 0.6) & (0.3, 0.5) & 0.5 \end{matrix}] \end{matrix}$ (9) $\begin{matrix} L_{23} - L_{3} \\ = [\begin{matrix} 0.5 & (0.2, 0.3) & (0.4, 0.5) \\ (0.7, 0.8) & 0.5 & (0.3, 0.5) \\ (0.5, 0.6) & (0.5, 0.7) & 0.5 \end{matrix}] \end{matrix}$ (10)

Furthermore, the fuzzy approximate judgement matrixes of the above four matrixes are calculated by the method in [31].

$F_{L 1} = [\begin{matrix} 0.5 & 0.889 & 0.778 \\ 0.111 & 0.5 & 0.5 \\ 0.222 & 0.5 & 0.5 \end{matrix}]$ (11) $F_{L 21} = [\begin{matrix} 0.5 & 0.889 & 0.625 & 0.667 \\ 0.111 & 0.5 & 0.111 & 0.222 \\ 0.375 & 0.889 & 0.5 & 0.714 \\ 0.333 & 0.778 & 0.286 & 0.5 \end{matrix}]$ (12) $F_{L 22} = [\begin{matrix} 0.5 & 0.5 & 0.571 \\ 0.5 & 0.5 & 0.625 \\ 0.429 & 0.375 & 0.5 \end{matrix}]$ (13) $F_{L 23} = [\begin{matrix} 0.5 & 0.222 & 0.444 \\ 0.778 & 0.5 & 0.375 \\ 0.556 & 0.625 & 0.5 \end{matrix}]$ (14)

We examine the consistency of the above-mentioned four intuitionistic fuzzy complementary judgement matrixes from (11) to (14) according to algorithm 1 in [31]. Here, we set λ = 0.4 and d = 0.1. We obtain satisfactory consistency matrixes from (15–18) by a consistency examination and adjustment of these matrixes. $F_{L 1}^{'} = [\begin{matrix} 0.5000 & 0.5591 & 0.5659 \\ 0.4819 & 0.5000 & 0.4655 \\ 0.4925 & 0.4669 & 0.5000 \end{matrix}]$ (15) $F_{L 21}^{'} = [\begin{matrix} 0.5000 & 0.5244 & 0.5333 & 0.5282 \\ 0.4527 & 0.5000 & 0.4475 & 0.4453 \\ 0.5276 & 0.5177 & 0.5000 & 0.5223 \\ 0.5011 & 0.4824 & 0.4988 & 0.5000 \end{matrix}]$ (16) $F_{L 22}^{'} = [\begin{matrix} 0.5000 & 0.5116 & 0.5025 \\ 0.5156 & 0.5000 & 0.5206 \\ 0.4649 & 0.4586 & 0.5000 \end{matrix}]$ (17) $F_{L 23}^{'} = [\begin{matrix} 0.5000 & 0.4523 & 0.4563 \\ 0.5104 & 0.5000 & 0.5144 \\ 0.4980 & 0.5353 & 0.5000 \end{matrix}]$ (18)

Finally, the weight vector is calculated. After obtaining a satisfactory consistency matrix, we can obtain the weight, W = (0.375, 0.3158, 0.3198), of all attributes of the second layer to the target layer according to formula (1) in [31].

Similarly, we can obtain the final weight of every attribute in the third sub-criteria layer to that in the upper one sub-criteria layer. The final weight is shown in Table 1.

Table 1

The grade of criteria importance based on intuitionistic fuzzy AHP

	L ₂₁	L ₂₂	L ₂₃
Attribute	0.375	0.3158	0.3198	Weight
L ₃₁	0.2643	0.3380	0.3029	0.3027
L ₃₂	0.2242	0.0000	0.0000	0.0841
L ₃₃	0.2613	0.3454	0.0000	0.2071
L ₃₄	0.0000	0.3078	0.0000	0.0972
L ₃₅	0.0000	0.0000	0.3416	0.1092
L ₃₆	0.2470	0.0000	0.3444	0.2028

After obtaining evaluation value (h) and the importance (g) of every attribute, every node can calculate the total evaluation value of cluster head according to cluster head evaluation hierarchy structure in Fig. 1. Firstly, the evaluation value of node starts from leaf node. We can obtain the evaluation value of the upper layer by calculating fuzzy integral [27] with the evaluation value and the weight. In the same way, we can obtain the final cluster head evaluation by the evaluation value of the upper layer and the weight of this layer [28, 29].

3.2 Establishing the cluster topology

In this subsection, we introduce the clustering algorithm of DGUCF to establish the unequal cluster topology. First, the cluster heads are selected using a timing broadcasting mechanism. The key problem is mapping the evaluation value and the broadcasting time. Second, the competition radius of a cluster head is determined. Finally, the cluster topology is formed in 3.2.3.

3.2.1 Mapping the evaluation value and the broadcasting time of cluster head

Cluster head broadcasting, in this paper, is similar to the timing broadcasting mechanism in [9]. The evaluation value of a cluster head maps on the timeline of triggering the cluster head selection. If the evaluation value of a cluster head is higher, the time of broadcasting message HEAD _ MASSAGE is faster. The formula of broadcasting the competition message is (19) as follows. $T_{i} = δ \times (1 - {CS}_{i}) \times T_{C}$ (19)

Here, δ is a random number between 0.9 and 1; CS_i is the evaluation value of a cluster head by a fuzzy integral; T_C is the pre-set time of forming a cluster.

3.2.2 The competition radius of a cluster head

We can obtain the message time of a competing cluster head from the above sub-section. When the time of a time-meter reaches that of a competing cluster head T_i, the node broadcasts a competition CH message HEAD _ MASSAGE with the competition radius of the cluster head to the network. The other nodes receiving this message HEAD _ MASSAGE will give up the chance to compete for the cluster head role.

There is heterogeneous initial energy to the nodes in heterogeneous WSNs. When the energy consumption of a node is the same, the node with the lower initial energy will die early. Nodes with a higher initial energy can take on more tasks. Hence, the nodes consider not only the distance to the BS but also the initial energy when establishing the unequal cluster topology. However, the unequal cluster topology cannot be effectively established if these two factors are only considered in the non-uniform deployment network. Specially, the distance between these nodes and BS is similar, and the residual energy is similar. Therefore, we propose a novel competition radius of the cluster head as formula (20). $R_{iC} = [1 - \underset{(1)}{\underset{︸}{\begin{matrix} (α \frac{d_{max} - d_{(i, Sink)}}{d_{max} - d_{min}} \\ + β (\frac{E_{max} - E_{ir}}{E_{max}}) + γ \frac{n_{i}}{N}) \end{matrix}}}] R_{0}$ (20)

Here, d_max and d_min are the maximal and minimal distances to BS; d_(i, Sink) denotes the distance between node i and BS; E_max is the maximal residual energy in the neighbour nodes of node i; E_ir indicates the current residual energy of node i; n_i is the total number of neighbour nodes; N is the total number of nodes in the networks; R₀ denotes the initial competition radius of cluster head. α, β and γ is the weight between 0 to 1, and the sum of parameters α, β, γ equals one. Otherwise, it is noteworthy that the value of R_ic is affected by three factors. In one existing case, if the result of (1) in formula (20) is greater than one, the value of R_ic is negative. This does not meet the real situation. Accordingly, the value range of α, β and γ is α + β + γ ≤ 1. According to the application to a real network, the value of α, β and γ can be adjusted dynamically so that the network lifetime can be prolonged with the optimal value of α, β and γ.

Formula (20) indicates that the size of a cluster is larger when the cluster head is farther from the BS, there is more residual energy and there are fewer neighbour nodes. There are three reasons that explain this case.

3.2.3 Forming cluster topology

When the time of a network is T_C, the competition phase of a cluster head is over. The network will form the cluster topology. In this phase, every common node sends a joining message JOIN _ MESSAGE to its nearest cluster head. This message includes the node ID and the current residual energy. According to the massage JOIN _ MESSAGE, every cluster head makes a time slot for every member node, i.e., makes a time table of TDMA. The time table shows when a member node will transmit data to its cluster head. After generating the time table, the cluster head will broadcast to its member nodes. If the role of a node is neither a cluster head nor a member node, the node will become a cluster head. Hence, the unequal cluster topology has been formed.

3.3 Data transmission

In this subsection, a new method for data transmission between cluster heads is explored, which to relay the aggregated data in proportional to two factors which are the distance between CHs and the residual energy. The proposed method tries to balance the relaying load of the CHs in order to balance their energy consumption.

In the data transmission phase, every member node senses and gathers data from the sensing area periodically. These data are transmitted to the cluster head according to a TDMA time slot, which avoids message collision within the same cluster. After receiving all data from the member nodes, the cluster head aggregates these data and transmits the data package to BS. This process can be divided into the following two phases: intra-cluster and inter-cluster communication.

If the distance between the cluster head and BS is not beyond the threshold, the data are directly transmitted to BS. If the distance is beyond the threshold, the data are transmitted to BS by the proposed routing algorithm. The specific process is as follows. First, every cluster head broadcasts a message CH _ RELAY _ MESSAGE with a broadcasting radius of 2R_ic. The message includes the cluster head ID, the current residual energy and the distance between the cluster head and BS. After receiving the massage, if the distance between the neighbour cluster head and BS is smaller than the one between CH_i and BS, CH_i will calculate these distances between CH_i and its neighbour cluster head CH_j (j = 1, 2 … L, L is the number of a neighbour cluster head). The table of neighbour cluster head is established.

CH_i calculates the cost function DRC according to formula (21). The data are transmitted to CH_j according to the ratio of the distributing data. $DRC (i, j) = 1 - \frac{E_{t (i, j)}}{E_{jr}}$ (21)

Here, E_t(i, j) is the energy consumption of transmitting data from CH_i to CH_j; E_jr is the current residual energy of CH_j.

Assuming the data amount of CH_i is M_i, the amount according to the ratio of distributing data in formula (22) is as follows.

$\begin{matrix} DRC (i, 1) : DRC (i, 2) : DRC (i, 3) : \\ DRC (i, 4) \dots DRC (i, L) \end{matrix}$ (22)

The amount of data transmitted from CH_i to its neighbour cluster head 1 is M_i × DRC (i, 1) and the amount transmitted to its neighbour cluster head 1 is M_i × DRC (i, 2), etc.

The core concept of the proposed routing algorithm is that the neighbour cluster head with more residual energy relays a higher ratio of the data amount. The proposed routing algorithm can effectively balance the energy consumption of inter-clusters and prolong the network lifetime. The routing algorithm according to the ratio of distributing data is shown in Fig. 2. For example, we assume that there are nine cluster heads from CH₁ to CH₉. The data amount of CH₁ is M₁ = 5000 bits, and its neighbour cluster heads are CH₃ and CH₄. According to formula (26), the data amount of CH₁ to obtain the ratio of distributing data is DRC (CH₁, CH₃) =0.4 and DRC (CH₁, CH₄) =0.3. Hence, the data amount transmitted from CH₁ to CH₃ is 5000 bits×0.4 = 2000 bits, and the amount transmitted from CH₁ to CH₄ is 5000 bits×0.3 = 1500 bits. In this way, the data is transmitted to BS.

Fig.2

Routing algorithm according to the ratio of distributing data.

4 Simulations

All the simulation experiments were performed on a computer with a 2.2 GHz dual core Intel Pentium processor and 2 GB RAM. We implemented DGUCF and three other data gathering protocols using Matlab R2013(b) under equivalent experimental conditions. The three other data gathering protocols are EEUC [6], IDUC [25], and FAHP [9]. All other experimental parameters are listed in Table 2.

Table 2
Simulation parameters

Parameters Value

BS location (250, 100)

Initial energy 1–3 J

The size of data packet 4000 bits

Communication radius 25 m

E _elec 50 nj/bit

ɛ _fs 10 pJ/(bit×m²)

E _SEN 0.5 J/bit

E _COM 5 nJ/(bit×signal)

d ₀ 80 m

Parameters	Value
BS location	(250, 100)
Initial energy	1–3 J
The size of data packet	4000 bits
Communication radius	25 m
E _elec	50 nj/bit
ɛ _fs	10 pJ/(bit×m²)
E _SEN	0.5 J/bit
E _COM	5 nJ/(bit×signal)
d ₀	80 m

The description of the two simulation scenarios is as follows.

100 nodes are uniformly deployed in a 200 m×200 m area.

100 nodes are non-uniformly deployed in a 200 m×200 m area.

The distribution of the sensor nodes is shown in Fig. 3. It is noteworthy that FAHP is another clustering algorithm with multiple criteria decision making. The core concept of FAHP is to establish a cluster topology of the same size. As FAHP is only a clustering algorithm, it does not involve inter-cluster communication. To compare with the other data gathering protocols, the inter-cluster communication of FAHP adopts the same routing algorithm used in this paper.

Fig.3

Simulation scenarios.

4.1 Establishing the cluster topology

The performance of establishing the cluster topology mainly demonstrates that DGUCF is fit for a network with a non-uniform distribution of nodes. Establishing the unequal cluster topology mainly depends on the competition radius of the cluster head. To prove this feature, the experiments are divided into two groups. In one group, the parameters of the competition radius R_C are set at α = 0.5, β = 0.5, γ = 0, and R₀ = 120m; In the other group, the parameters are set at α = 0.3, β = 0.3, γ = 0.4, and R₀ = 120m. In Scenario II, DGUCF with two different group parameters is executed. The two different unequal cluster topologies are shown in Fig. 4.

Fig.4

Establishing the cluster topology.

As shown in Fig. 4, the unequal clustering topology with parameter II is more reasonable than that without the non-uniform distribution of nodes. This result can be explained by the fact that the competition radius of the cluster head with parameter II can be adjusted according to the number of neighbour nodes. The size of a cluster with more neighbour nodes is smaller, and the size of a cluster with fewer neighbour nodes is larger. Therefore, the imbalance of energy consumption is alleviated due to the non-uniform distribution of the nodes.

4.2 Distribution of the cluster heads

Establishing the unequal cluster topology mainly depends on the competition radius of the cluster heads in formula (20). According to (20), the parameters R₀, α, β and γ are very important for determining the value of the competition radius. In formula (20), R₀ is the initial maximal competition radius of the cluster head. α, β and γ are the weight of the competition cluster heads. The three factors are the distance to BS, residual energy and the number of neighbour nodes. As mentioned in the above section, the value of (20) is always greater than 0. Hence, the sum of α, β and γ must be lower than 1. DGUCF is executed in scenario II to determine the effect of parameters R₀, α, β, γ on the cluster topology. In this scenario, R₀ is set from 10 to 200 m, and the values of α, β, γ are divided into three groups. They are (α= 0, β= 0, γ= 0), (α= 0.2, β= 0.2, γ= 0.3) and (α= 0.3, β= 0.3, γ= 0.4). The purpose of the three groups is as follows. The first group only considers the relationship between R₀ and the number of cluster heads; In the case of α + β + γ < 1 in the second group, the number of cluster heads is generated by DGUCF; In the case of α + β + γ = 1 in the third group, the number of cluster heads is generated by DGUCF.

As shown in Fig. 5, regardless of the combination of α, β and γ, the number of cluster heads decreases as the initial competition radius R₀ increases. When the value of α, β, γ is (0.3, 0.3, 0.4), the number of cluster heads generated by DGUCF is more than (0.2, 0.2, 0.3). When the value of α, β, γ is (0.2, 0.2, 0.3), the number of cluster heads generated by DGUCF is more than (0, 0, 0). This result can be explained by the fact that the value of R_C is determined by R₀, α, β and γ. When the value of R₀ is fixed, if the sum of α, β and γ is greater, the value of R_C is lower, and the number of cluster heads generated by DGUCF is greater. When the sum of α, β and γ is fixed, if the value of R_C is greater, the value of R_C is greater, and the number of cluster heads generated by DGUCF is lower.

Fig.5

Relationship between the initial value of R₀ and the number of cluster head.

Next, the stability of the clustering algorithm of DGUCF is investigated. After the nodes are deployed, they are static in WSN. The number of cluster heads generated by a stable cluster algorithm should be relatively centralized, which can decrease the energy consumption. In [17], an approximate optimal formula for the number of cluster heads is theoretically obtained in a single-hop network. Thus, the number of cluster heads generated by the clustering algorithm is an expected value. This value is the optimal number of cluster heads in a scenario. In scenarios I and II, four data gathering protocols, EECU, FAHP, IDUC and DGUCF, are executed.

Figure 6 presents the number of cluster heads for the four data gathering protocols in the two scenarios. As shown in Fig. 6, the number of cluster heads for the four protocols is relatively concentrated. However, the two clustering algorithms of IDUC and DGUCF are more stable than EECU and FAHP. In scenarios I, the stability of FAHP is better than that of DGUCF. Since FAHP is an equal size clustering algorithm, it is fit for the network scenario of a uniform distribution of nodes. In scenarios II, the stability of DGUCF is better than that of FAHP.

Fig.6

Amount distribution of cluster head.

4.3 Network lifetime

In this subsection, we compare the network lifetime of the four protocols in the two experimental scenarios. The key parameters of DGUCF are set at R_max= 140 m, α= 0.3, β= 0.3, γ= 0.4. IDUC is another unequal clustering protocol in the heterogeneous network. Its key parameters are set at α= 0.3, β= 0.3, γ= 0.4, ω= 0.5 R_max= 140 m.

As shown in Fig. 7, we can observe that the network lifetime of DGUCF and IDUC is longer than that of EEUC and FAHP. This result can be explained by the fact that it is reasonable that DGUCF and IDUC involve the number of neighbour nodes when establishing the cluster topology. The network lifetime of DGUCF is longer than that of IDUC. The energy consumption of DGUCF is more balanced because DGUCF adopts not only multiple criteria decision making to compete the cluster heads but also the ratio distribution routing algorithm to communicate with the inter-clusters.

Fig.7

Comparison of the network lifetime of the four protocols in scenario I.

As shown in Fig. 8, the number of live nodes for the four data gathering protocols decreases linearly after the first dead node occurs. Otherwise, we can observe that the network lifetime of DGUCF is superior to that of the other three protocols.

Fig.8

Comparison of the network lifetime of the four protocols in scenario II.

This result can be explained by the fact that DGUCF is an unequal clustering topology. In the clustering, DGUCF adopts an intuitionistic fuzzy analytic process and a hierarchical fuzzy integral to select cluster heads synthetically. A routing algorithm according to the ratio of distributed data is used to balance the energy consumption of inter-cluster communication. Accordingly, the network lifetime of DGUCF is obviously superior to that of the other three protocols.

4.4 Data traffic

In this subsection, we compare the receiving data traffic of Sink for the four data gathering protocols in the two scenarios. As shown in Figs. 9 and 10, we can observe that the total number of packets delivered to the BS is cumulative and steadily increases as the simulation time increases. The gradual decrease in the slope of the curve towards the end of the simulation is due to the nodes failing as the simulation progresses in time.

Fig.9

Data traffic comparison of the four protocols in scenario I.

As shown in Fig. 9, we can observe that receiving data traffic of UCDGAMCD and FAHP is more than that of the other two algorithms at the same round in scenario I. This result can be explained by the fact that the factor of transmission reliability is considered during cluster head competition in UCDGAMCD and FAHP. Furthermore, our results show that the total number of packets reaches nearly 20,000 in UCDGAMCD while this value cannot reach 20,000 in the other algorithms. The cluster heads are chosen randomly in LEACH, while the other three protocols evenly distribute the cluster heads. This reduces the energy loss due to transmission for the nodes expected to transmit frequently, thereby delivering the same amount of data with less energy dissipation.

As shown in Fig. 10, we can observe that the receiving data traffic of UCDGAMCD is more than that of the other three algorithms at the same round in scenario II. This result can be explained by the fact that the factor of transmission reliability is considered during cluster head competition in UCDGAMCD. The competition radius of a cluster head is very fit for scenario II, the unequal cluster topology can be reasonably established and Sink receives more data traffic.

Fig.10

Data traffic comparison of the four protocols in scenario II.

Comparing Figs. 9 and 10, we can observe that the data traffic of UCDGAMCD and IDUC inscenario II is more than that in scenario I. This can be explained by the fact that an unequal cluster topology is reasonably established in scenario II. However, the data transmission reliability of IDUC is not considered; thus, the receiving data traffic is less than that of UCDGAMCD. Because data transmission reliability is considered by FAHP, its data traffic in the initial rounds is more than that of IDUC. As the simulation time increases, the receiving data traffic of FAHP gradually decreases due to FAHP with a uniform cluster topology.

5 Conclusion

In this paper, DGUCF adopts intuitionistic fuzzy analytic process and hierarchical fuzzy integral to select cluster heads. Then, we propose a new cluster head competition radius, which is fit for the network scenarios with energy heterogeneous and nodes non-uniform distribution. In succession, to balance energy consumption, each CH transmits the data packets over the next hop CHs in proportional to their residual energy. Finally, the simulation results demonstrate that DGUCF protocol can balance energy consumption effectively and the network lifetime can be prolonged.

Although DGUCF demonstrates the better performance, other application scenarios such as fault-tolerance, mobility wireless sensor networks are not involved. In future, we plan to propose unequal data gathering protocol in a more comprehensive network scenario.

Footnotes

Acknowledgments

This research is supported by the National Nature Science Foundation of China (61170169, 61170168).

References

Mahmood

M.A.

, Seah

W.K.G.

and Welch

, Reliability in wireless sensor networks: A survey and challenges ahead, Computer Networks79 (2015), 166–187.

Yao

, Cao

and Vasilakos

A.V.

, EDAL: An energy-efficient, delay-aware, and lifetime-balancing data collection protocol for heterogeneous wireless sensor networks, IEEE/ACM Transactions on Networking23(3) (2015), 810–823.

Sabor

, Abo-Zahhad

, Sasaki

and Ahmed

S.M.

, An unequal multi-hop balanced immune clustering protocol for wireless sensor networks, Applied Soft Computing43 (2016), 372–389.

Liu

A.F.

, Zhang

P.H.

and Chen

Z.G.

, Theoretical analysis of the lifetime and energy hole in cluster based wireless sensor networks, Journal of Parallel and Distributed Computing71(10) (2011), 1327–1355.

Soro

and Heinzelman

, Prolonging the lifetime of wireless sensor networks via unequal clustering, Proceedings of the 5th International Workshop on Algorithms for Wireless, Mobile, Washington DC, USA, IEEE Computer Society, 2005, pp. 1–8.

Chen

G.H.

, Li

C.F.

, Ye

, et al., An unequal cluster-based routing protocol in wireless sensor networks, Wireless Networks15(2) (2009), 193–207.

Lai

W.K.

, Fan

C.S.

and Lin

L.Y.

, Arranging cluster sizes and transmission ranges for wireless sensor networks, Information Sciences183(1) (2012), 117–131.

Dutta

, Gupta

and Das

M.K.

, Low-energy adaptive unequal clustering protocol using fuzzy c-means in wireless sensor networks, Wireless Personal Communications79(2) (2014), 1187–1209.

Gao

, Jin

R.C.

, Song

J.Y.

and Wang

, Energy-efficient cluster head selection scheme based on multiple criteria decision making for wireless sensor networks, Wireless Personal Communications63(4) (2012), 871–894.

10.

Mardani

, Jusoh

and Zavadskas

E.K.

, Fuzzy multiple criteria decision-making techniques and applications–Two decades review from 1994 to 2014, Expert Systems with Applications42(8) (2015), 4126–4148.

11.

Atanassov

K.T.

, Intuitionistic fuzzy sets, Fuzzy Sets and Systems20(1) (1986), 87–96.

12.

Bagci

and Yazici

, An energy aware fuzzy approach to unequal clustering in wireless sensor networks, Applied Soft Computing13(4) (2013), 1741–1749.

13.

Mao

, Zhao

, Zhou

and Ye

, An improved fuzzy unequal clustering algorithm for wireless sensor network, Mobile Networks and Applications18(2) (2013), 206–214.

14.

Hematkhah

and Kavian

Y.S.

, DCPVP: Distributed clustering protocol using voting and priority for wireless sensor networks, Sensors15(3) (2015), 5763–5782.

15.

Heinzelman

W.R.

, Kulik

and Balakrishnan

, Adaptive protocols for information dissemination in wireless sensor networks, Proceedings of the 5th Annual ACM/IEEE International Conference on Mobile Computing and Networking, ACM, 1999, pp. 174–185.

16.

Heinzelman

W.B.

, Chandrakasan

A.P.

and Balakrishnan

, An application-specific protocol architecture for wireless microsensor networks, IEEE Transactions on Wireless Communications1(4) (2002), 660–670.

17.

Bandyopadhyay

and Coyle

E.J.

, An energy efficient hierarchical clustering algorithm for wireless sensor networks, INFOCOM 2003, 23th Annual Joint Conference of the IEEE Computer and Communications, IEEE Societies, 3, 2003, pp. 1713–1723.

18.

Younis

and Fahmy

, HEED: A hybrid, energy-efficient distributed clustering approach for ad hoc sensor networks, IEEE Transactions on Mobile Computing3(4) (2004), 366–379.

19.

Kumar

, Aseri

T.C.

and Patel

R.B.

, EEHC: Energy efficient heterogeneous clustered scheme for wireless sensor networks, Computer Communications32(4) (2009), 662–667.

20.

Logambigai

and Kannan

, Fuzzy logic based unequal clustering for wireless sensor networks, Wireless Networks22(3) (2016), 945–957.

21.

Nayak

and Devulapalli

, A fuzzy logic-based clustering algorithm for wsn to extend the network lifetime, IEEE Sensors Journal16(1) (2016), 137–144.

22.

Sert

S.A.

, Bagci

and Yazici

, MOFCA: Multi-objective fuzzy clustering algorithm for wireless sensor networks, Applied Soft Computing30 (2015), 151–165.

23.

Jain

and Ramana Reddy

B.V.

, A novel method of modeling wireless sensor network using fuzzy gragh and energy efficient fuzzy based k-hop clustering algorithm, Wireless Personal Communications82 (2015), 157–181.

24.

, Fu

, Cai

and Vuran

M.C.

, A reliable energy-efficient multi-level routing algorithm for wireless sensor networks using Fuzzy petri nets, Sensors11 (2011), 3381–3400.

25.

Chen

, Li

, Gu

, Yu

and Huang

, IDUC: An improved distributed unequal clustering protocol for wireless sensor networks, International Journal of Distributed Sensor Networks (2015), 12. Article ID415702.

26.

Liu

, Zhang

and Chen

, On mitigating hotspots to maximize network lifetime in multi-hop wireless sensor network with guaranteed transport delay and reliability, Peer-to-Peer Networking and Applications7(3) (2014), 255–273.

27.

Tzeng

, Yang

, Lin

and Chen

, Hierarchical MADM with fuzzy integral for evaluating enterprise intranet web sites, Information Sciences169(34) (2005), 409–426.

28.

Zhang

, Ma

and Xu

, A new fuzzy MCDM method based on trapezoidal fuzzy AHP and hierarchical fuzzy integral, Fuzzy Systems and Knowledge Discovery, SpringerBerlin Heidelberg, 2005, pp. 466–474.

29.

Sugeno

, Theory of fuzzy integrals and its applications, Ph.D. Dissertation, Tokyo Institute of technology, 1974.

30.

, Yu

and Lv

, ECDC: An energy and coverage-aware distributed clustering protocol for wireless sensor networks, Computers & Electrical Engineering40(2) (2014), 384–398.

31.

Zhang

, Li

and Wang

, AHP under the intuitionistic fuzzy environment, Proceedings of 8th International Conference on Fuzzy Systems and Knowledge Discovery, Shanghai, 2011, pp. 583–587.

Energy-efficient data gathering protocol in unequal clustered WSN utilizing fuzzy multiple criteria decision making

Abstract

Keywords

1 Introduction

2 Related work

3 DGUCF protocol

3.1 Cluster head selection

3.2.1 Mapping the evaluation value and the broadcasting time of cluster head

3.3 Data transmission

Table 2 Simulation parameters Parameters Value BS location (250, 100) Initial energy 1–3 J The size of data packet 4000 bits Communication radius 25 m E elec 50 nj/bit ɛ fs 10 pJ/(bit×m2) E SEN 0.5 J/bit E COM 5 nJ/(bit×signal) d 0 80 m

Footnotes

Acknowledgments

References

Table 2
Simulation parameters

Parameters Value

BS location (250, 100)

Initial energy 1–3 J

The size of data packet 4000 bits

Communication radius 25 m

E _elec 50 nj/bit

ɛ _fs 10 pJ/(bit×m²)

E _SEN 0.5 J/bit

E _COM 5 nJ/(bit×signal)

d ₀ 80 m