A Fire Development Prediction Model for Large Complex Connected Tunnels Based on Simulation

Abstract The construction of large interchange tunnels addresses urban traffic congestion but introduces complex operational safety challenges, with fire being among the most severe hazards. To mitigate tunnel fire risks and enhance emergency response capabilities, this study proposes a rapid prediction model for tunnel fire development based on Long Short-Term Memory (LSTM) and Fully Connected Neural Network (FCNN) models. The model requires only basic inputs—including heat release rate (HRR), fire location, longitudinal ventilation, and tunnel geometry—to quickly forecast key fire parameters, such as ceiling temperature, temperature at 2 m below the ceiling, and visibility over time. Given the high cost and data scarcity of full-scale tunnel fire experiments, an extensive fire scenario database was generated using Fire Dynamics Simulator (FDS) simulations under varying fire locations, HRRs, and ventilation conditions. Time-series data extracted from these simulations were used to train and validate the neural networks. Comparative analysis showed that the LSTM model achieved a coefficient of determination (R 2 ) of 0.8952, outperforming the FCNN model (R 2 = 0.8721) in capturing temporal fire dynamics. The predictive capability of the neural network model was further validated against theoretical ceiling temperature models derived from the FDS data, confirming its feasibility for practical application in guiding fire rescue and ventilation strategies. It should be noted that the model's performance is inherently dependent on the accuracy and scope of the underlying simulation data, and its reliability may decrease when extrapolating beyond the trained parameter ranges.