Generative AI-Driven Adaptive Consistency Maintenance for Distributed Simulation Systems

Conventional consistency maintenance methods in distributed simulations degrade under unstable networks due to fixed rules and limited adaptability. To address this, we propose an AI-driven, two-layer adaptive consistency framework. In Tier-1, lightweight diagnostic tools use rule-based checking for immediate anomaly detection and initiate a protective "Soft-pause."In Tier-2, an edge-deployed Large Language Model acts as a decision engine, analyzing multi-dimensional network metrics to generate adaptive repair strategies via structured JSON. Extensive experiments under dynamic network degradation demonstrate our framework achieves a 98.6% Detection Accuracy and over 93% Consistency Maintenance Rate, significantly outperforming traditional baselines in preventing system crashes despite seconds-level repair latency.