The main concern of clustering approaches for mobile wireless sensor networks (WSNs) is to prolong the battery life of the individual sensors and the network lifetime. For a successful clustering approach the need of a powerful mechanism to safely elect a cluster head remains a challenging task in many research works that take into account the mobility of the network. The approach based on the computing of the weight of each node in the network is one of the proposed techniques to deal with this problem. In this paper, we propose an energy efficient and safe weighted clustering algorithm (ES-WCA) for mobile WSNs using a combination of five metrics. Among these metrics lies the behavioral level metric which promotes a safe choice of a cluster head in the sense where this last one will never be a malicious node. Moreover, the highlight of our work is summarized in a comprehensive strategy for monitoring the network, in order to detect and remove the malicious nodes. We use simulation study to demonstrate the performance of the proposed algorithm.
After the success of theoretical research contributions in previous decade, wireless sensor networks (WSNs) have become now a reality [
Clustering formation of WSNs composed of 150 sensor nodes deployed in a 570 m × 555 m space area with a radio range = 100 m.
In this paper, we propose an energy efficient and safe weighted clustering algorithm for mobile WSNs using a combination of the above metrics to which we added a behavioral level metric. The latter metric is decisive and allows the proposed clustering algorithm to avoid any malicious node in the neighborhood to become a CH, even if the remaining metrics are in its favor. The election of CHs is carried out using weights of neighboring nodes which are computed based on selected metrics. So this strategy ensures the election of legitimate CHs with high weights. The preliminary results obtained through simulation study demonstrate the effectiveness of our algorithm in terms of the number of equilibrate clusters and the number of reaffiliations, compared to WCA (Weighted Clustering Algorithm) [
We can enumerate the contributions of our paper as follows: maintaining stable clustering structure and offering a better performance in terms of the number of reaffiliations using the proposed algorithm ES-WCA (Energy Efficient and Safe Weighted Clustering Algorithm); detecting common routing problems and attacks in clustered WSNs, based on behavior level; showing clearly the interest of the routing protocols in energy saving and therefore maximizing the lifetime of the global network.
The remaining part of this paper is organized as follows. Section
In this section, we outline some approaches of clustering used in ad hoc networks and WSNs. Research studies on clustering in ad hoc networks involve surveyed works on clustering algorithms [
In this paper, the proposed approach focuses around strategy of distributed resolution which enables us to generate a reduced number of balanced and homogeneous clusters in order to minimize the energy consumption of the entire network and prolong sensors lifetime. The introduction of a new metric (the behavioral level metric) promotes a safe choice of a cluster head in the sense where this last one will never be a malicious node. Thus, the highlight of our work is summarized in a comprehensive strategy for monitoring the network, in order to detect and remove the malicious nodes.
The fact that WSNs include limited energy resources (batteries) due mainly to their small size, our algorithm shows clearly the interest of the routing protocols in energy saving which therefore maximize the lifetime of the network by coupling it with AODV and then DSDV protocols [
The typical attacks in WSNs include Sinkhole attack, Black Hole attack, Hello Flood attack, and Node Outage which are the most common network layer attacks on WSNs [
Sinkhole attack is one of the most devastating ones: it is very hard to protect against [
In this attack, malicious nodes advertise very short paths (sometimes zero-cost paths) to every other node, forming routing black holes within the network [
Many routing protocols use “Hello” broadcast messages to announce themselves to their neighbor nodes. The nodes that receive this message assume that source nodes are within range and add source nodes to their neighbor list. The Hello Flood attacks can be caused by a node which broadcasts a Hello packet with very high power, so that a large number of nodes even far away in the network choose it as the parent node [
If a node acts as an intermediary, an aggregation point, or a cluster head, what happens if the node stops working? Protocols used by the WSNs must be robust enough to mitigate the effects of failures by providing alternate routes [
This section introduces the different metrics used for cluster head election by focusing on behavior level metric.
The behavioral level of a node
Behavior level
Our objective is to have stable clusters. So, we have to elect nodes with low relative mobility as CHs. To characterize the instantaneous nodal mobility, we use a simple heuristic mechanism as presented in the formula below (
In our previous paper [
(a) Clustering mechanism in mobile WSNs before moving nodes and (b) after moving nodes 1, 5, and 4.
This is likely to reduce node detachments and enhance cluster stability. For each node
The residual energy of a node
It represents the number of
For each node, we must calculate its weight
These thresholds are arbitrarily selected or they depend on the topology of the network. Thus, if their values depend on the topology of the network, they are calculated as follows according to [
We denote AVG by the average cardinal of the groups with one jump of all the nodes of the network:
In this section, we first present some assumptions of the proposed algorithm: Energy Efficient and Safe Weighted Clustering algorithm (ES-WCA). Then we present in detail an extended version of ES-WCA [
This paper is based on the following assumptions. The network formed by the nodes and the links can be represented by an undirected graph All sensor nodes are deployed randomly in a 2-dimension (2D) plane. A node interacts with its one-hop neighbors directly and with other nodes via intermediate nodes using multihop packet forwarding based on a routing protocol such as ad hoc on demand distance vector [ The radio coverage of sensor nodes is a circular region centered on this node with radius Two sensor nodes cannot be deployed in exactly the same position All sensor nodes are identical or homogeneous. For example, they have the same radio coverage radius Each node can determine its position at any moment in a 2D space. Each cluster is monitored by only one CH. Each CM communicates directly with its CH for the transmission of security metrics. A CH communicates directly with the base station for the transmission of security information and possible alerts.
The ES-WCA algorithm that we present below is based on the ideas proposed by Chatterjee et al. [
ES-WCA uses three types of messages in the setup phase (Algorithm
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15)
(15) (16) (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) (27) (28) (29) (30) (31) (32)
Procedure of affiliation of node “U” to a cluster.
The node which has the greatest weight begins the procedure by broadcasting CH message to their 1-hop neighbors to confirm its role as a leader of the cluster. The neighbors confirm their role as being member nodes by broadcasting a JOINmsg message. In the case when nodes have the same maximum weight, the CH is chosen by using the best parameters ordered by their importance. If all parameters of nodes are equal, the choice is random.
ES-WCA uses four types of messages in the reaffiliation phase (Algorithm
(1) (2) (3) (4) (5) (6) (7)
(9)
(11) CH → CH → Size = Size + 1; (12) (13) (14) (15) Go to (2); (16)
Procedure of reaffiliation of node “U” to a cluster.
With the help of 3 figures (Figures
Values of the various criteria of normal nodes.
Ids |
|
|
|
|
|
|
---|---|---|---|---|---|---|
1 | 0.86 | 3842.12 | 3 | 1.15 | 1.20 | 769.632 |
4 | 0.81 | 4832.54 | 5 | 2.30 | 0.30 | 968.133 |
5 | 0.88 | 4053.25 | 3 | 1.30 | 0.55 | 811.829 |
6 | 0.85 | 4620.43 | 0 | 0.00 | 0.20 | 924.361 |
8 | 0.81 | 4816.80 | 4 | 1.05 | 1.40 | 964.753 |
10 | 0.95 | 3650.25 | 2 | 0.55 | 0.10 | 730.805 |
11 | 0.91 | 4819.60 | 1 | 0.70 | 2.20 | 964.753 |
Weights of neighbors.
Ids | 1 | 4 | 5 | 6 | 8 | 10 | 11 |
|
|||||||
1 | 769.632 |
|
|
||||
4 |
|
811.829 | 964.753 | ||||
5 |
|
811.829 | 730.805 | ||||
6 |
|
||||||
8 | 769.632 |
|
|||||
10 |
|
811.829 | 730.805 | ||||
11 | 769.632 |
|
Topology of the network.
Identification of clusters node.
The final identification of clusters.
At the end of this example, we obtain a network of four clusters (as shown in Figure
Final cluster structure (reaffiliation phase).
There are five situations that require the maintenance of clusters: battery depletion of a node, behavior level of a node less than or equal 0.3, adding, moving, or deleting a node.
In all of these cases, if a node
Monitoring in WSNs can be both local and global. The local monitoring can be with respect to a node and the global monitoring can be with respect to the network, but in sensor networks, for detecting some types of errors and security anomalies, the local monitoring would be insufficient [
Monitoring phase architecture.
Monitoring phase.
Number of packets sent by Number of packets received by node Delay between the arrivals of two consecutive packets is Energy consumption: the energy consumed by the node
In the initial deployment, each CM in cluster “ Each state contains the following information: If (state then node Otherwise, no information is sent to the CH. The message received by CH If a sensor node The behavior level of sensor node The “rate” is fixed on the basis of the nature of the application. For example, if it is fault tolerant or not. In our case, we took rate = 0.1.
For each node where with For a normal node, Here, In our case, the first interval is used for the training data set of After modeling a normal behavior model for each sensor node, the behaviors of all nodes are sent to the base station for further analysis. We then compute the deviation When the deviation
The punishing algorithm is presented in Algorithm
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
This section presents the implementation of the proposed approach using the Borland C++ language and the analysis of the obtained results.
We try to complete the theoretical study by implementing our own wireless sensor network simulator “Mercury.” On the other hand, a bit of simulators for WSNs such as TOSSIM [
To evaluate our ES-WCA algorithm, we have performed extensive simulation experiments. This section provides our experimental results and discussions. In all the experiments,
The performance of the proposed ES-WCA algorithm is measured by calculating (i) the number of clusters, (ii) number of reaffiliations, (iii) choice of ES-WCA with AODV or DSDV, and (iiii) detection of misbehavior nodes and the nature of attacks during the distributed monitoring process.
In our experiments, the values of weighting factors used in the weight calculation are as follows:
Figure
Average number of clusters versus transmission range (
Figure
Average number of clusters versus number nodes (
Figure
Average number of clusters versus transmission range ES-WCA and WCA.
Figure
Average number of reaffiliations.
Remaining energy per node using ES-WCA.
We consider that the network will be inoperative when the nodes of the neighborhood of the sink exhaust their energy as exemplified. In Figure
Network lifetime depending on number of nodes using ES-WCA.
To illustrate the effect of abnormal behavior in the network, in our experiments we propagated 200 nodes with 5 malicious nodes. The cases of the malicious nodes will pass from a normal node with a yellow color to an abnormal node with a blue color, to a suspicious node with a grey color, and lastly, to a malicious node with a black color. All the cases of the CMs are discovered by their CH. Malicious CHs are disclosed by the base station.
Figure
Detection of the nature of attacks.
IDs | Packets_Sent | Packets_Received | Attack |
---|---|---|---|
41 | (19, 13) | (16, 14) | Node Outage |
71 | (24, 152) | (20, 34) | Hello Flood |
162 | (15, 8) | (22, 112) | Sinkhole |
181 | (16, 179) | (26, 42) | Hello Flood |
190 | (58, 32) | (50, 51) | Black Hole |
(a) Graph connectivity of 200 nodes. (b) Network after clustering formation.
(a) Sensors with a blue color are abnormal but not malicious. (b) The grey sensors have a suspect behavior. (c) The sensors with a black color are compromised and are exhibiting malicious behavior.
Behavior level of some sensors (moves frequently).
Behavior level of some sensors before and after attacks.
In this paper, we have presented a new algorithm called “ES-WCA” for promoting the self-organization of mobile sensor networks. This algorithm is fully decentralized and aims at creating a virtual topology with the purpose to minimize frequent reelection of the cluster head (CH) and avoid overall restructuring of the entire network. Simulations result attest of the outperformance of our algorithm compared to WCA and DWCA in every sense. It yields a low number of clusters and it preserves the network structure better than WCA and DWCA by reducing the number of reaffiliations. The proposed algorithm selects the most robust and safe CHs with the responsibility of monitoring the nodes in their clusters and maintaining clusters locally. Our third algorithm analyses and detects specific misbehavior in the WSNs. The results show that in scenarios in which mobile WSNs are with a low density or with a small size, the choice of ES-WCA with AODV is comparable to ES-WCA with DSDV to show clearly the interest of the routing protocols in energy saving. However, the difference in favor between ES-WCA and AODV becomes very important in case of a high node density. This is due to the tremendous overheads incurred by ES-WCA with DSDV when exchanging routing tables and exchanging routing control packets. Future work includes considering further the concept of redundancy by using the “sleep” and “wakeup” mechanism in case of node failure, providing in-network processing by aggregating correlated data in order to reduce both the energy consumption and the congestion issue.
The authors declare that there is no conflict of interests regarding the publication of this paper.
The authors are grateful to the anonymous referees for their insightful comments and valuable suggestions, which greatly improved the quality of the paper.