Dynamic Threat Assessment for Small-UAV Intrusions into Nuclear Power Plants via Cost-Sensitive Sequential Three-Way Decisions

The physical protection system (PPS) of a nuclear power plant faces a high-consequence and low-tolerance risk profile in low-altitude small–unmanned aerial vehicle (UAV) intrusion scenarios. The core task in such scenarios is to minimize the risk of underestimating high-threat targets under conditions of incomplete multi-source evidence, fluctuating credibility of identity information, and limited response time. This paper proposes a dynamic threat-assessment method for nuclear power plant PPSs. The method organizes the outputs of radar, radio-frequency detection/Remote ID, electro-optical verification, and information analysis and processing into progressive evidence granularities. It first performs coarse assessment using low-cost information; then acquires identity, localization, and electro-optical evidence on demand for uncertain samples; and finally performs safety-side commitment when the time-to-intrusion is insufficient. Single-target intrusion simulations show that compared with fixed three-way decision using all evidence, the proposed method reduces the average online test cost from 2.900 to 0.788, reduces the asymmetric risk from 0.938 to 0.666, and increases the recall of high-threat targets from 0.836 to 0.962. These results indicate that the proposed method reduces unnecessary high-cost evidence acquisition without weakening safety conservatism and is suitable for dynamic threat assessment, capability evaluation, and vulnerability diagnosis under nuclear power plant PPS constraints.