Predicting critical nodes of Opportunistic Sensor Network (OSN) can help us not only to improve network performance but also to decrease the cost in network maintenance. However, existing ways of predicting critical nodes in static network are not suitable for OSN. In this paper, the conceptions of critical nodes, region contribution, and cut-vertex in multiregion OSN are defined. We propose an approach to predict critical node for OSN, which is based on multiple attribute decision making (MADM). It takes RC to present the dependence of regions on Ferry nodes. TOPSIS algorithm is employed to find out Ferry node with maximum comprehensive contribution, which is a critical node. The experimental results show that, in different scenarios, this approach can predict the critical nodes of OSN better.
In Opportunistic Sensor Network (OSN), the critical nodes are very important to keep normal operation of networks. In practical applications, if the critical nodes can be predicted, the network could be optimized according to the attributes of critical nodes, which helps improving the robustness of the network. In network maintenance, maintainers can focus on monitoring the status of critical nodes so that the failures of the network could be resolved immediately, which can dramatically reduce the time and the cost of network maintenance. Therefore, predicting critical nodes of OSN has great significance.
OSN is a kind of Wireless Sensor Networks. It perceives the surrounding environment by sensor nodes and transports messages by the meeting opportunities of Ferry nodes. Hence, it has the characteristics of Mobile Opportunity Network [
Corley and Sha [
With the research above, it is always one-sided to evaluate the network through a single evaluation index. Considering the influence of node degree, node closeness, node betweenness, equivalent topology, and neighbor lists, Hu et al. [
As a dynamic network, current static prediction methods of critical nodes are not appropriate for OSN. Depending upon the researches above, in this paper, the stage contribution and the region contribution are proposed to evaluate the node importance. Then an algorithm is designed which is based on the multiple attribute decision making (MADM) to predict the critical nodes of OSN.
The monitoring areas of application scenarios like environmental monitoring are very large. Therefore, the maintainers tend to monitor the key regions instead of the whole network. In OSN, the messages of the network are collected through the communication opportunities supported by mobile nodes. As shown in Figure
OSN scenario.
Scenario
Scenario model
In Figure
In this paper, the following assumptions are made: In our research, each region is abstracted as a “super node” called region node. Regardless of the Ferry nodes’ memory, it is assumed that Ferry node can collect all the messages from each node it meets. Regions and Ferry nodes and Sink nodes in the network have unique identity information. The network has a time synchronization mechanism.
In Figure
Define Opportunistic Sensor Networks as
OSN is a dynamic network which transfers data by the “Store-Carry-Forward” mechanism. The traditional parameters such as node degree [
In OSN, Ferry nodes are transport mediums. So their job is to transport messages between Sink nodes and region nodes. In order to accurately estimate each Ferry node’s importance to the network, the effect of Ferry nodes on OSN must be considered properly. With intensive analysis of OSN’s routing mechanism, the region messages’ life cycle can be divided into three stages as shown in Figure
OSN message transmission.
In the first stage, Ferry nodes receive network messages from regions and then carry them out. In the second stage, the network messages will be forwarded among Ferry nodes. At last, Ferry nodes transport messages to the Sink nodes. These three stages can not only depict the message propagation of OSN clearly but also show the important role of Ferry nodes obviously.
Define a time slice as
Define a time slice as
Define a time slice as
Define a time slice as
The region contribution can reflect both the Ferry nodes’ contributions to regions and the dependence of the regions on Ferry nodes. It means that the bigger region contribution the node has, the higher possibility leading to the network split the node possesses and the node is more likely to be a critical node. If the region contribution from node
According to the researches above, we can infer that the node is a cut-vertex of the network when the region contribution equals 1 and it must be a critical node.
According to the research mentioned above, the critical node prediction method for OSN can be described as the following steps.
Calculate each Ferry node’s region contributions in order to determine whether the network has cut-vertexes or not. If there are no cut-vertexes in the network, go to Step
Find out a node which most likely leads to the network split and it must be a critical node. The region contribution shows the dependence of regions on Ferry nodes. We can learn that the higher region contribution the node has, the higher risk of network split it will have. Based on the theory above, we first take each Ferry node as a single evaluation scheme. Then, the TOPSIS method is applied to evaluate the comprehensive region contribution of Ferry nodes.
It is meaningless to predict such a dynamic network like OSN by calculating region contributions within a single time slice such as
We assume an OSN with
Denote the maximum of
For comparison, the decision matrix could be optimized by the following normalization processing:
Then the decision matrix could be updated to
Due to the different importance of different regions to the whole network, weight is assigned for each evaluation index to make the algorithm more universal. We denote the
According to matrix
We denote the distance from every solution
Then we calculate and sort the ideal solutions similarity degree
According to the TOPSIS method, the node with maximum
It is assumed that OSN has At first, denote the length of time as Construct weighted normalized matrix Determine the positive ideal solution Calculate the distance from every solution Calculate the similarity degree between each solution and ideal solution by ( Repeat the above steps and denote the length of prediction time as
As is shown in Figures
Scenario A.
Simulation
Model
Scenario B.
Simulation
Model
Scenario C.
Simulation
Model
Scenario D.
Simulation
Model
In Figure
In Figure
In some more complex situations like Scenario C and Scenario D, in Figure
In Figure
We have made experiments 100 times for each scenario. Table
Experiment results.
Scenario | CN | Appearance possibility |
|
Result | |||||
---|---|---|---|---|---|---|---|---|---|
fa | fb | fc | fd | fe | fg | ||||
Scenario A | fa | 0.33 | 0.04 | 0 | 0.06 | 0.03 | 0.06 | 0.33 | fa |
Scenario B | fd | 0.31 | 0 | 0.35 | 0.52 | 0.02 | 0 | 0.52 | fd |
Scenario C | fa | 0.44 | 0.02 | 0.24 | 0.01 | 0 | 0 | 0.44 | fa |
Scenario D | fc, fd, fe | 0.22 | 0.17 | 0.52 | 0.52 | 0.52 | — | 0.52 | fc, fd, fe |
As shown in Table
Considering the dynamic of OSN, this paper proposed a MADM based method to predict critical nodes. First, the region contribution is introduced to present the dependency of regions on Ferry nodes. Then, the comprehensive region contributions are estimated by the MADM method. At last, the experimental results show that, for different OSN scenarios, our method can predict the critical nodes of the network effectively.
The authors declare no competing interests.
This work is supported in part by grants from the National Natural Science Foundation of China (nos. 61262020, 61363015, 61501218, and 61501217).