LLM-Driven Semantic Consistency Analysis for TRDP Implementations in Train Networks

Abstract

To address the difficulty of detecting silent inconsistencies in Train Real-time Data Protocol implementations, we propose TRDP-Guard, an automated protocol consistency checking architecture based on large language model (LLM)-driven semantic analysis. TRDP-Guard extracts structured rules from the IEC 61375 standard, localizes relevant implementation logic via LLM guided program slicing, and performs semantic consistency analysis between rules and code to identify latent deviations. Experiments on real-world TRDP implementations show that TRDP-Guard effectively detects representative defects such as timer mismanagement, PULL/PUSH misuse, and missing field validation, achieving 86.4% precision and 95.0% recall, significantly outperforming existing static analysis tools. Furthermore, we model three classes of consistency breaking attacks timing, scheduling, and semantic confusion attacks and design corresponding defense mechanisms. Simulation results demonstrate that the proposed defenses effectively mitigate protocol inconsistency-induced threats, enhancing the robustness of train communication systems.