A Cost-Effective Data-driven Approach to Flashover Prediction across Diverse Residential Layouts for Enhanced Firefighters Situational Awareness

Published

December 30, 2024

Author(s)

Linhao Fan, Hongqiang Fang, Tianshui Liang, Wai Cheong Tam, Qixing Zhang

Abstract

This paper presents gFlashNet, a generic flashover prediction model, designed to address the limitations of existing models that are restricted to specific residential building layouts. The aim of this research is to improve the scalability and adaptability of flashover prediction models, which is crucial for enhancing fire safety in buildings. By representing the spatial positions of sensors as a graph network, gFlashNet can be applied to various building layouts without requiring any model structure modifications. A graph attention mechanism is integrated to strengthen the model's ability to capture sensor connectivity during fire events. Additionally, transfer learning is employed to reduce development costs by enabling the pre-trained model to be fine-tuned on new layouts using a smaller dataset. The result shows that gFlashNet achieves high prediction accuracy for new layouts with significantly less data, reducing data requirements compared to traditional approaches. This work contributes a novel, cost-effective approach for developing generalizable fire safety models, with significant potential for real-time flashover prediction across diverse residential layouts.

Citation

Journal of Building Engineering

Pub Type

Journals

Download Paper

Local Download

Keywords

Flashover Prediction, Spatio-Temporal Graph Neural Network, Model Scalability, Transfer Learning, Smart Firefighting

Citation

Fan, L. , FANG, H. , Liang, T. , Tam, W. and Zhang, Q. (2024), A Cost-Effective Data-driven Approach to Flashover Prediction across Diverse Residential Layouts for Enhanced Firefighters Situational Awareness, Journal of Building Engineering, [online], https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=958283 (Accessed January 2, 2025)