Lightweight and Efficient Distributed Cooperative Intrusion Detection System for Intelligent Swarms

Abstract

Intrusion Detection Systems (IDSs) have been wildly used in various environments to actively detect internal and external attacks with high accuracy. Unfortunately, the traditional IDSs cannot distinguish new or unknown attacks from abnormal behaviors effectively, and that makes them infeasible to protect the emerging dynamic open information systems. Subsequently, artificial intelligence (AI) algorithms are introduced into IDSs to support the recognization of unexplored malicious behaviors. However, most of the existing AI-driven IDSs are not able to be directly applied to intelligent swarm scenarios, which are typically employed to aggregate heterogeneous or homogeneous elements (e.g., autonomous vehicles, drones) to solve complex problems that the individual members cannot deal with, due to the characteristics of mobility and complexity of intelligent elements. Therefore, in this paper, we propose a lightweight and efficient distributed cooperative IDS (DCIDS) for intelligent swarms. On one hand, to efficiently detect the malicious behaviors among swarm elements, we design a collaborative detection model which utilizes multi-dimension features including the swarm elements' position, storage-computing resource consuming levels, network traffics, et al. On the other hand, to predict the movement trends and detect attacks of resource-limited swarm elements, we construct a concrete DCIDS scheme by employing the Kalyan Filter algorithm and Long Short Term Memory Network (LSTM) algorithm. Furthermore, our experimental results demonstrate that the proposed DCIDS scheme outperforms the previous IDS schemes on attack detection/classification accuracy and efficiency in intelligent swarm environments and also achieves an accuracy of 98.00%.