Intelligent prediction method for fire temperature fields in underground exhibition spaces of high-intensity urban areas

The detection of fire temperature fields in underground exhibition spaces has become a critical issue for fire evacuation planning. This study aims to elucidate the influence mechanisms of spatial characteristics on fire temperature fields and innovatively proposes a temperature field prediction method based on distributed fiber optic temperature sensors. The results indicate that different exhibition space layouts and heights significantly affect the temperature field distribution of key fire prevention planes, with the impact of space heights being more significant than that of layouts. Based on this, the Fire Dynamics Simulator (FDS) platform was used to simulate the fire temperature fields under different conditions, constructing fire temperature databases. Through a comparative selection of multiple algorithms, it was found that the RF prediction model with 41 input features performs best in terms of accuracy and applicability, with the Mean Absolute Errors (MAE) values of 1.09°C and 0.52°C for fire rooms and no-fire rooms, respectively, and the Mean Absolute Percentage Errors (MAPE) values of 1.11% and 0.68%, respectively. Finally, a multi-feature generalization test was conducted to verify the model's good generalization performance in new scenes. This study provides technical support with application prospects for fire evacuation in underground exhibition spaces. • A prediction method for fire temperature field in exhibition space is proposed • Space feature has a significant impact on temperature distribution • Derive temperature mapping relationship based on sensor temperature and space feature • Fire temperature prediction model error based on RF algorithm is 1.11% • Prediction model demonstrates good generalization across different space features