A novel unmanned aerial vehicle driven real-time situation awareness for fire accidents in chemical tank farms

Abstract

A large number of flammable hazardous materials are stored in chemical tank farms, where fire-induced domino accidents can be easily triggered. In this study, a novel real-time fire situation awareness (FSA) approach based on UAV is proposed to capture spatio-temporal evolution characteristics and predict development trends of fire accidents. Firstly, fire images are acquired by UAV, and the key parameters of fire are extracted in real time based on YOLOv8 network. Then, the thermal radiation and impact on surrounding equipment are predicted by combining LSTM network, solid flame model and improved probit model. The proposed method is verified by small-scale tank fire experiments, which demonstrate its superiority in terms of physical consistency and prediction accuracy. The results show that the mean absolute percentage error (MAPE) of fire parameter extraction is not higher than 5.43%, the MAPE of thermal radiation prediction is not higher than 25%, and the dynamic time to failure ($dttf$) for the model tank at different location is predicted. This work has the potential to provide a novel solution for real-time assessment of fire size and trend prediction to support firefighting, emergency rescue and decision making in fire accident scenarios.