Rapid bidirectional prediction between physical field and key control parameters in tunnel fires

Abstract

Tunnel fire poses a serious threat to social public safety, and the losses they cause are often incalculable. The prediction of tunnel fires contributes to decision-making in rescue and firefighting, and helpfully reduces fire losses as much as possible. The financially expensive experiments and the time-consuming simulation slow down the pace of development in tunnel fire prediction. Moreover, numerical and experimental study is often unidirectional, with the characteristic of predicting less dimensional data through higher dimensional data. This work proposes a deep learning model (DLM) to instantly achieve bidirectional prediction between the full field information of tunnel fires and a small amount of key physical quantities. Under the designed data processing method, the DLM is trained by a big tunnel fire numerical database with various ventilation, thermal, and geometric conditions. The results show that the DLM can learn the physical fields data and the physical quantities data well with the increasing training epoch. In addition, the DLM performs the promising bidirectional prediction. From the symmetry comparison, the result shows the full physical fields are well predicted by the decoder part of DLM via four key physical quantities. The prediction of the key physical quantities is overall satisfactory, but the prediction accuracy of the tunnel inclination angle is relatively poor compared with the other quantities. The prediction accuracy of key physical parameters through the temperature field is better than through smoke visibility. The important parameters in practice, namely smoke layer distribution and smoke back-layering length are also predicted, and the R${}^2$ of 0.95 and 0.92 are respectively obtained. The bidirectional prediction system proposed in this work demonstrates the promising application for intuitive and rapid prediction of various information in tunnel fires, as well as for summaries of physical laws in tunnel fires.