Citation

Zhu L. Safety Sci. 2019; 118: 96-102.

Copyright

(Copyright © 2019, Elsevier Publishing)

DOI

10.1016/j.ssci.2019.05.012

PMID

unavailable

Abstract

The Sensing network attack connection has the characteristics of behavioral variability and complexity. It is not feasible to construct an abnormal intrusion detection model by using behavior mining technology based on traditional clustering. According to the characteristics of Sensing network attack behavior, this paper proposed a Sensing network attack detection algorithm based on improved multi-cluster. Firstly, it used improved spatial clustering algorithm to reduce spatial data feature dimensions and feature computational complexity during attack detection. Secondly, it used improved K-means clustering algorithm to classify spatial datasets. Ant colony optimization algorithm is used to detect whether there is Sinkhole attack in routing and generate alert information of sensor nodes. Finally, it used improved evidence accumulation clustering algorithm to calculate the difference distance between each isolated point and the cluster centroid and it also used matrix clustering algorithm to calculate the detection threshold and determine the attack behavior in the Sensing network. P2P trust model is used to calculate the trust value of each node in the list of suspect nodes, and the node whose trust value is lower than the preset threshold is regarded as the attack node. Through the attack detection experiment based on KDD 99 dataset, the comparison with the detection results of different algorithm s shows that the proposed algorithm has higher detection rate and lower false positive rate.

Language: en

Keywords

Ant colony optimization; Data fusion; Evidence accumulation; K-means clustering; Sensing network attack; Spatial clustering