Simple model combined with data assimilation to forecast buoyancy-driven ceiling jet propagation in tunnel fires

Abstract

Faster-than-real-time forecast for the ceiling jet head propagation is crucial for evacuation guidance and rescue operation in the early stages of tunnel fires. The ceiling jet head propagation is driven by buoyancy and thus significantly dominated by the heat transfer process in the head region. Simplified models have been established but proved challenging in forecasting complex dynamic processes. The cumulative deviations induced by the model limitations and the input parameter uncertainties lead to distorted forecast during the ceiling jet dynamic propagation process, thereby constraining the applicability of such models in practical scenarios. In this study, an assimilation method, the ensemble Kalman filter (EnKF), was combined with a simple theoretical model to forecast the ceiling jet propagation. Using the EnKF to assimilate the monitored local data, the key input parameters such as mass flow rate m˙ and the temperature of the ceiling jet head Ts are corrected in real time, thus alleviating model prediction deviation. Meanwhile, a heat loss rate correction coefficient λα, introduced to correct the heat loss rate during the ceiling jet propagation process, is dynamically estimated as well. This enables the model to characterize the heat transfer process more accurately, consequently ensuring faster-than-real-time forecasts of the subsequent propagation stage. Finally, the forecasting effectiveness of the proposed combined method was examined in a 1:6 reduced-size experimental tunnel fire scenario and full-scale simulated tunnel fire scenarios with various fire source locations, tunnel geometries, and suddenly changing HRR. In response to the demand of minimizing both the type and quantity of sensors in practical applications, a factor analysis was also performed to examine the robustness of this method. The results exhibited good adaptability to different fire scenarios with very limited monitored data.