Data-Driven Prediction Methods for Real-Time Indoor Fire Scenario Inferences

Abstract

High temperatures, toxic gases, and smoke resulting from indoor fires pose evident threats to the lives of both trapped individuals and firefighters. This study aims to predict indoor fire development effectively, facilitating rapid rescue decisions and minimizing casualties and property damage. A comprehensive database has been developed using Computational Fluid Dynamics (CFD) tools, primarily focused on basic fire scenarios. A total of 300 indoor fire scenarios have been simulated for different fire locations and severity levels. Using fire databases developed from simulation tools, artificial intelligence models have been developed to make spatial–temporal inferences on indoor temperature, CO concentration, and visibility. Detailed analysis has been conducted to optimize sensor system layouts while investigating the variations in prediction accuracy according to different prediction horizons. The research results show that, in combination with artificial intelligence models, the optimized sensor system can accurately predict temperature distribution, CO concentration, and visibility, achieving R2 values of 91%, 72%, and 83%, respectively, while reducing initial hardware costs. The research results confirm the potential of artificial intelligence in predicting indoor fire scenarios and providing practical guidelines for smart firefighting. However, it is important to note that this study has certain limitations, including the scope of fire scenarios, data availability, and model generalization and interpretability.