Risks Evaluation of Utility Tunnel for Gas Leakage, Fire and Explosion Disasters

Utility tunnels become more and more common in urban areas around world. However, utility tunnels pose several risks, including potential leakage of hazardous gas, fire, and explosion. Full-scale Utility tunnel tests on fire temperature distribution was carried out firstly to measure the temperature field variation in the tunnel. The fire detection and early alarming method was proposed based on multi-source data trained lightweight Convolutional Neural Networks (CNNs). Three Long Short-Term Memory (LSTM) classification models were developed to identify the fire source location, Heat Release Rate (HRR), and wind speed in different scenarios. A deep learning model combining LSTM and Transpose Convolutional Neural Networks (TCNN) was established for the spatial inference of indoor temperature distributions given limited temperature sensor data. Preliminary coupled disasters evaluation was also made for possible risks with CH4 gas leakage, fire and explosion.