Intrusion detection based on deep reinforcement learning model fusion for maritime meteorological sensor networks

Maritime meteorological sensor networks (MMSN) differ from traditional land-based networks, presenting new challenges for intrusion detection tasks. A satellite-based detection method for maritime meteorological sensor networks was designed using satellite communication technology. The network structure and characteristics of maritime meteorological sensor networks were analyzed in this method. Research was conducted on improving the detection performance of intrusion detection systems (IDS) from the perspectives of algorithms and loss functions. A maritime meteorological sensor network intrusion detection method based on the fusion of deep reinforcement learning models was proposed. Firstly, light gradient boosting machine (LightGBM), 1D conventional neural network (1D-CNN), and 2D conventional neural network (2D-CNN) classifiers with improved loss functions were established to comprehensively extract the temporal and spatial features of the intrusion detection data in maritime meteorological sensor networks. Secondly, a model fusion method was designed based on the stacking and averaging principles of model fusion technology. This method leveraged the strengths of the base classifiers and mitigated their weaknesses, thereby enhancing the overall system detection performance. Finally, simulation experiment results demonstrate that the proposed intrusion detection method can effectively improve the detection performance for a few types of attack data and enhance the robustness of the system.