Smart real-time forecast of transient tunnel fires by a dual-agent deep learning model

Abstract

Disastrous fire accidents occurred in the tunnel is fatal and destructive, which may pose great threats to the trapped person and firefighters. However, fire in a confined tunnel can develop rapidly and spread between vehicles, so it is difficult to forecast the fire development and possible catastrophic incidents. This work develops an intelligent model to predict the fire information, temperature distribution and critical events in real-time based on artificial intelligence algorithms. The numerical model is first validated by the full-scale tunnel fire test, and then a numerical database of 300 transient tunnel-fire scenarios is established under various initial fire locations, fire sizes, fire growth and spread rates, and ventilation conditions. The proposed dual-agent deep-learning model combining the Long Short-term Memory (LSTM) model and Transpose Convolution Neural Network (TCNN) is trained with the database. With the input data of on-site temperature sensors, the dual-agent model can forecast transient fire scenarios with changing location and size 30 s in advance. This study demonstrates the feasibility of the AI model in identifying and forecasting the rapid-changing fire scenarios inside a tunnel in smart firefighting practices.