A Data-Driven Danger Zone Estimation Method Based on Bayesian Inference for Utility Tunnel Fires and Experimental Verification

Abstract

The ongoing challenge is to find an effective and precise fire detection method for safety concerns of utility tunnels to predict the fire danger zone and take measures for firefighting and intervention promptly. A data-driven danger zone estimation method was established based on Bayesian inference. The probability distribution of the fire danger zone can be obtained by this method. In particular, the governing equation is a simplified physical model, in which only crucial parameters of the fire states are referred to, including the fire source location, the maximum temperature, and the attenuation coefficient. This task shows superiority because it can avoid additional workload to provide the forward database for Bayesian inference. A prototype experiment was conducted in the largest utility tunnel experimental platform in China to verify the validity of the proposed method. Results demonstrated that the fire danger zone could be estimated with high accuracy merely based on several sensor data, which are not limited to specific scenes. The temperature distribution of the whole tunnel could also be predicted according to the fire parameters. Moreover, analysis of the measurement noise and disturbance conditions shows the robustness of the proposed method. Low costs, low consumption, and universality make it have broad engineering application prospects.