전력전자 부하 지표와 물리내재 LSTM을 활용한 스푸핑 내성 차압 제어 = Spoofing-Resistant Differential Pressure Control using Power Electronics Load Indicators and Physics-Informed LSTM

다국어 초록 (Multilingual Abstract)

Maintaining room–hallway differential pressure (ΔP) is essential for infection control (negative-pressure isolation) and smoke backflow prevention (positively pressurized zones). Sensor spoofing—bias, drift, delay, and replay—can mislead controllers, causing unsafe over/under-pressure and excessive fan burden. This paper specifies a physics-informed LSTM (PI-LSTM) residual detector whose loss embeds mass balance and fan laws, and analyzes a lightweight confusion-index (CI) proxy aggregating fan power, pressure deviation, and flow imbalance. A small-chamber HVAC simulator generates physically credible synthetic datasets (door-leak toggling, low-frequency wind offsets, sensor noise, parameter uncertainty). On this baseline, a simple residual score exhibits clear separation between normal and attacked segments, with strong ROC/PR characteristics, while the CI consistently rises during attacks, summarizing increased system burden. As a simulation-only baseline study, this work provides a reproducible evaluation path and motivates future closed-loop CI-guided mitigation and power-electronics-derived stress-metric validation on hardware.