This study proposes a physics-informed surrogate framework for generating synthetic compartment fire time-series data in sensor-scarce building environments. The generator solves an energy-balance ODE with two-zone modeling, wall conduction, ventilation loss, and radiation loss. We validate 20 representative scenarios against real CFAST and FDS simulations, achieving MAE of 15.8°C against FDS volume-mean. Downstream ML benchmarks confirm data utility (R²=0.918 vs. R²=−346 for naive baselines).
목차
Abstract 1. 서론 2. 본론 2.1 시나리오 구성 2.2 에너지 보존 ODE 기반 온도 계산 3. 검증 실험 3.1 CFAST/FDS 비교 결과 3.2 ML 활용 검증 4. 결론 Acknowledgments References