LMD: a large-scale model-driven defense mechanism for cyber-resilient smart ports

Global maritime and port logistics systems face escalating cyber-physical vulnerabilities due to increasing digitization, geopolitical tensions, and climate disruptions. To address the limitations of static defense mechanisms in dynamic port environments, this paper proposes LMD—a Large-scale Model-driven Defense Mechanism integrating self-supervised representation learning, lightweight prefix-tuning, and cloud-edge-carrier collaborative moving-target defense. LMD leverages unlabeled maritime traffic to learn semantic behaviors through masked-reconstruction and context-prediction objectives, employs prefix vectors for rapid adaptation to emerging threats with minimal parameter updates, and implements a software-defined networking (SDN) based adaptive IP-hopping mechanism to isolate malicious traffic while maintaining low-latency communications. Extensive experiments across diverse attack scenarios demonstrate that LMD achieves a 6.99% higher detection accuracy, 8.21% improved normal communication capability, and 9.18% reduced CPU load compared to baseline methods, significantly enhancing cyber-resilience in smart port infrastructures. This framework offers a scalable, resource-efficient paradigm for safeguarding maritime communication networks against evolving threats.