OceanCL-IDS: A Continual Learning-based Intrusion Detection System for Ocean-going Ship-Satellite Communication Network

Deep learning-based intrusion detection has become an effective approach to ensure the security of communication network traffic in scenarios such as the Internet of Things (IoT) and the Internet of Vehicles (IoV). However, there is relatively little research on intrusion detection for ship-satellite communication network in marine scenarios. In addition, due to the inherent catastrophic forgetting issue of deep learning models, vanilla intrusion detection methods can only recognize previously learnt attack types, but cannot learn new attack types in real time. In this paper, an ocean-going ship-satellite communication network (OSSCN) intrusion detection system (OceanCL-IDS) is proposed, which can effectively alleviate the communication interference caused by the harsh maritime communication environment. Furthermore, a continual learning algorithm GEM based on replay strategy is introduced into the system, which is capable of continual learning new attack features while retaining the capacity to detect previously learnt attack. Finally, to validate the efficacy of the proposed methods, a dataset with attack labels and maritime communication features is constructed by collecting long-term satellite network traffic data of real ocean-going ships. Extensive experimental evidence demonstrates that our method successfully beats existing continual learning algorithms on a variety of metrics.