Physical modeling and machine learning of ceiling maximum temperature rise induced by tandem heat sources with unequal heat release rates in a natural ventilation tunnel

Abstract

The present work involved experimental studies of physical modeling as well as machine learning prediction of the characteristics of ceiling maximum temperature induced by adjacent heat sources with unequal fire heat release rates in a tunnel. The following are considered as variables in this study: separation distances between the adjacent heat sources, and unequal heat release rates between upstream fire sources (UFS) and downstream fire sources (DFS). It was found that, with the increasing separation distance between adjacent fires, the ceiling maximum temperature gradually decreases for the given total heat release rate (Q˙TFS). Meanwhile, with the increase in the ratio (Q˙UFS/Q˙TFS) of the upstream fire heat release rate (Q˙UFS) to the total heat release rate, the ceiling maximum temperature rise of the tunnel presents nonmonotonic changes: first a decrease and then an increase. From the dimensional analysis of adjacent tandem fires, a prediction model of ceiling maximum temperature rise induced by different separation distances and Q˙UFS/Q˙TFS ratios was proposed. The model has a good correlation with the experimental and previous research results. Furthermore, to perform the calculation quickly and facilitate rapid engineering design, we use machine learning methods to derive the maximum temperature rise of the tunnel ceiling. The fire heat release rate is predicted using a machine learning algorithm. The three parameters (Q˙UFS/Q˙TFS, ΔTmax, Hd) were paired with the maximum temperature rise of the ceiling and trained using the GRU neural network model. The results show that the machine learning algorithms has a good agreement with the verification experimental value, cmparisons with established physical models are also discussed. The research results can serve as a reference for an intelligent tunnel firefighting system.