Risk Prognosis and Fire Prediction of Urban Utility Tunnels Using a Hybrid SSA‐LSTM Method

ABSTRACT This paper presents a prediction model for the fire trend prediction in urban utility tunnels by integrating the sparrow search algorithm (SSA) with the long short‐term memory (LSTM) network. An improved LSTM is developed to effectively capture the dynamic nature of fire evolution. The SSA is employed to optimize the parameters of the LSTM, with its global optimization capability enhanced through the incorporation of four meta‐heuristic optimization methods. To comprehensively evaluate the model's effectiveness, comparative experiments were conducted against the Support Vector Machine (SVM), Random Forest (RF), Gated Recurrent Unit (GRU), and Transformer models, demonstrating the superiority of the improved SSA‐LSTM in multiple evaluation metrics. The advanced SSA‐LSTM model is then used to predict fire severity based on fire intensity. The applicability and effectiveness of the proposed model are validated through a practical fire experiment. Comparative analysis with existing approaches indicates that the proposed model achieves an approximately 15% improvement in prediction accuracy. In addition, the model shows potential for broader applications in dynamic fire trend identification and critical point warning systems.