In the complex operational environment of nuclear power plants, traditional digital alarm systems often suffer from issues such as information overload and key alarm omission, primarily caused by the “keyhole effect.” To address these challenges, this study proposes and evaluates an intelligent alarm management system that integrates multi-level flow modeling (MFM) and advanced inference techniques. The system's performance was validated through rigorous experimental testing, demonstrating its superiority in response time, alarm compression rate, and false compression rate. The results indicate that the intelligent alarm system effectively reduces redundant alarm information, enhances key alarm identification accuracy, and improves operator response efficiency. Specifically, the system achieves a more than 15% increase in operational accuracy compared to traditional alarm systems. By optimizing decision-support capabilities, the proposed intelligent alarm system significantly enhances the stability and safety of nuclear power plant operations. Furthermore, it establishes a foundation for subsequent human-factor experiments, underscoring its substantial engineering value and practical significance.