Архитектура MAS-XAI для интерпретируемого управления транспортными потоками

Abstract

This study develops a framework for integrating Multi-Agent System (MAS) with Explainable AI (XAI) to address interpretability challenges in traffic management. The proposed three-layer framework implements local XAI modules (LIME, SHAP) at the agent level, explanation aggregation at the coordinator level, and human-machine interaction at the user level. Framework is expected to enhance operators trust and enable more effective traffic management in complex urban environments.