Intrusion Detection for Maritime Transportation Systems With Batch Federated Aggregation

As a fast-growing and promising technology, Internet of Things (IoT) significantly promotes the informationization and intelligentization of Maritime Transportation System (MTS). The massive data collected during the voyage is usually disposed of with the assistance of cloud or edge computing, which imposes serious cyber security threats. For multifarious cyber-attacks, Intrusion Detection System (IDS) is one of the efficient mechanisms to prevent IoT devices from network intrusion. However, most of the methods based on deep learning train their models in a centralized manner, which needs uploading all data to the central server for training, increasing the risk of privacy disclosure. In this paper, we consider the characteristics of IoT-based MTS and propose a CNN-MLP based model for intrusion detection which is trained through Federated Learning, named FedBatch. Federated Learning keeps the model training local and only updates the global model through the exchange of model parameters, preserving the privacy of local data on vessels. First, the characteristics of the communication between different vessels are discussed to model the federated learning process during the voyage. Then, the lightweight local model constructed by Convolutional Neural Network (CNN) and Multi-Layer Perception (MLP) is designed to save on computing and storage overhead. Moreover, to mitigate the straggler problem during the federated learning in MTS, we proposed an adaptive aggregation method, named Batch Federated Aggregation, which suppresses the oscillations of model parameters during federated learning. Finally, the simulation results on the NSL-KDD dataset demonstrate the effectiveness and efficiency of FedBatch.