arXiv:2610.16770v2 Announce Type: replace Abstract: Next-generation wireless networks are expected to support high data rates and reliable connectivity while meeting requirements for energy efficiency, resource management, and communication security. Digital Twin (DT) technology represents expected network behavior and enables the actual network state to be assessed against it. Integrating DT technology with reinforcement learning (RL) supports adaptive resource allocation and security-aware decision-making. This paper presents a simulation-based DT reference and a belief-guided Q-learning (BGQL) method for joint jammer identification and beam-power control in a site-specific 28-GHz network model. Receiver-level residuals between attack-free and attacked configurations are combined with candidate spatial jamming footprints and accumulated into a normalized suspicion vector. BGQL incorporates this vector into its state, reward, and exploratory beam-selection rule. Compared with SARSA, Greedy, and Random across 30 paired Monte Carlo runs per configuration, with known jammer counts, BGQL achieves link-rate scores of 437.4-564.2 Mbit/s, rate-to-device-power scores of 60.37-76.79 Mbit/J, and localization recall of 100% for one jammer and 78%-81% for two jammers. BGQL requires 80-130 s per run versus 5-14 s for SARSA. These findings provide a matched-model simulation proof of concept for DT-referenced belief-guided learning for security-aware beam-power control in 6G networks.