Cyber Digital Twin-Enabled Security Attack-Defense Target Range for Vocational Training: Integrating RTK with Lightweight Federated Learning

Abstract

The digitalization of vocational education has exposed e-commerce simulation platforms to sophisticated cyber threats, while conventional cybersecurity training suffers from static scenarios and absence of intelligent adversaries. This paper proposes a Cyber Digital Twin (CDT)-enabled attack-defense target range that integrates three synergistic innovations: (1) a hybrid twin architecture synchronizing containerized IT/OT assets with physical IoT devices; (2) the Red Team Knife (RTK) toolchain augmented with Large Language Model (LLM)-driven adaptive attack agents for dynamic Cyber Kill Chain simulation; and (3) a Lightweight Federated Learning framework (Lightweight-Fed-NIDS) employing structured pruning for real-time edge inference. A zero-trust middleware ensures secure cross-domain access with sub-10ms latency. Experimental validation on Raspberry Pi 4B hardware demonstrates 99.2% detection accuracy with 128ms inference latency, while pedagogical trials show 57% improvement in real-time response capabilities compared to conventional CTF training. This work establishes a scalable paradigm for immersive cybersecurity education in resource-constrained environments.