Dynamic Security Configuration in China Railway Cloud Center: A Combination of Mobile Target Defense and Q-Learning

The China Railway Cloud Center, as critical infrastructure, faces escalating cyber threats that demand proactive defense strategies beyond static mechanisms. This paper proposes a dynamic security configuration framework that combines Moving Target Defense (MTD) with Q-learning. We first model the cloud application as a stack of configurable layers (e.g., OS, Middleware) and formalize vulnerabilities and attackers using Common Vulnerability Scoring System (CVSS) metrics. A transition reward function is then derived, integrating attack probability, security entropy, switching cost, and CVSS-based impact. Finally, the configuration switching problem is formulated as a Markov Decision Process and solved using Q-learning to find the optimal policy. Simulation results demonstrate that the Q-learning strategy converges to configurations yielding higher cumulative rewards and lower costs compared to random, periodic, and greedy baselines, effectively reducing the simulated attack success rate from 34% (random) to 12%. This work provides an adaptive, learning-driven defense framework for railway critical information infrastructure, offering a foundation for automated security orchestration.