Detection of malicious nodes in drone ad-hoc network based on supervised learning and clustering algorithms

Abstract

Multi-drone swarm has been widely used in disaster monitoring, mapping and remote sensing, national defense military and other fields, and has become a research hotspot in recent years. Due to the openness of its operating environment, attackers can invade the control system to capture drone, and then carry out data attacks such as tamper attack, drop attack and replay attack in drone ad-hoc network, which causes a great threat to the security of drone network. Existing malicious nodes detection algorithms are not efficient when applied to drone ad-hoc network, for the following reasons: (1) The malicious node detection algorithms based on reputation usually adopt a static threshold to determine whether a node is malicious, which is inefficient in dynamic drone network. (2) Mutual cooperation based malicious node detection algorithms rely on the high meeting probability of nodes. In order to solve the above problems, we propose a Malicious Drones Detection Algorithm(MDA) based on supervised learning and clustering algorithms. The ground station calculates the reputation value of each routing path according to the received packets from different source nodes, and then evaluates the reputation value of drones with linear regression algorithm. Finally, gaussian clustering algorithm is used to cluster drones and find out malicious drones. Experiments were conducted in indoor and outdoor drone network. The experimental results indicate that the accuracy of MDA outperforms the existing methods by 10% 20%. And in the case of fewer malicious nodes, the accuracy can reach more than 90%, and the error rate is less than 10%.