On the Use of Machine Learning Models to Forecast Flashover Occurrence in a Compartment
Jun Wang, Wai Cheong Tam, Paul A. Reneke, Richard D. Peacock, Thomas G. Cleary, Eugene Yujun Fu, Grace Ngai, Hong va Leong
This paper presents a study to examine the potential use of machine learning algorithms to build a model to forecast the likelihood of flashover occurrence for a single-floor multi-room compartment. Synthetic temperature data for heat detectors from different rooms are generated using Fire Data Generator. More than 1000 simulation cases are considered and a total of 8 million data points are utilized for model development. The development of P-Flash (Prediction model for Flashover occurrence) is presented. Two special treatments: sequence segmentation and learning from fitting, are proposed to overcome the temperature limitation of heat detectors in real-life fire scenarios and to enhance prediction capabilities to forecast the future likelihood of flashover occurrence even with situations where there is no temperature data from all detectors. Experimental evaluation shows that P-Flash offers reliable prediction. The model performance is approximately 83.2 % and 81.7 %, respectively, for current and future likelihood of flashover occurrence, considering heat detector failure at 150 ̊C. Results demonstrate that P-Flash, a new data-driven model, is feasible to provide fire fighters real-time, trustworthy, and actionable information to enhance situation awareness, operational effectiveness, and safety for firefighting.
17th International Conference on Automatic Fire Detection (AUBE 20) & Suppression, Detection and
Signaling Research and Applications Conference (SUPDET 2020)