Vol 77 No 2 77205

Abstract

As the development of intelligent ships rapidly progresses, the importance of maritime network security escalates. Intrusion Detection Systems (IDS) serve as a pivotal defense mechanism against cyberattacks targeting these vessels. To augment the efficacy of IDS in detecting anomalous network traffic, this paper introduces a novel intrusion detection method utilizing both Transformer and Kolmogorov-Arnold Networks (KAN), designated as Transformer-KAN. This approach effectively mitigates the limitations of traditional intrusion detection algorithms, which include inadequate feature extraction capabilities and the excessive parameterization required to delineate complex patterns. The proposed method employs a Transformer encoder to discern long-range dependencies within input sequences, while KAN layers enhance the model’s ability to approximate complex patterns through non-linear transformations. This integrative strategy ensures elevated accuracy for the intrusion detection algorithm and proves particularly beneficial in intelligent ship environments, where training data may be sparse or computational resources limited. Experimental results confirm that the Transformer-KAN model attains exceptionally high accuracy in both binary and five-class classification tasks, outperforming the capabilities of standalone Transformer and KAN models. Comparative analyses with conventional algorithms—such as Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), Deep Belief Networks (DBN), and Support Vector Machines (SVM)-further substantiate the superior effectiveness of the proposed method. This research establishes a new theoretical framework and provides a direction for the practical implementation of network security measures in contemporary intelligent ships.