A Deep Learning-based Approach for Unsafe Area Prediction in Building Fire Evacuation

Published

August 10, 2025

Author(s)

Hongqiang Fang, Botao Zhang, Wai Cheong Tam, Chendi Yang, S.M. Lo

Abstract

Dynamic directional exit signs (DDES), also known as smart exit signs, are specifically developed to provide real-time guidance to evacuees during building fire emergencies. However, the existing research lacks adequate focus on effectively predicting unsafe areas in such situations. This insufficiency hampers the ability of DDES systems to incorporate dynamic fire condition awareness seamlessly into route planning algorithms, thereby restricting their applicability in fire emergencies. To address this gap, this paper introduces a novel approach that utilizes deep learning (DL) methods for dynamically predicting unsafe areas during building fire emergencies. By leveraging DL methods--Bidirectional-GRU-based encoder-decoder (BiGRU-ED), the proposed approach establishes a mapping between sequential data of on-site temperature observations and future states of fire characteristics. This enables the identification of potential hazards in building floor areas that may become unsafe zones in advance, which allows for adequate time to plan safer evacuation routes and control DDES systems accordingly. The results of a case study demonstrate the feasibility and effectiveness of the proposed approach, highlighting its potential for improving smart control of DDES systems in building fire emergencies.

Citation

Journal of Building Engineering

Pub Type

Journals

Download Paper

Local Download

Keywords

fire safety, dynamic evacuation, deep learning, GRU, fire exit signs

Citation

FANG, H. , Zhang, B. , Tam, W. , Yang, C. and Lo, S. (2025), A Deep Learning-based Approach for Unsafe Area Prediction in Building Fire Evacuation, Journal of Building Engineering, [online], https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=957437 (Accessed March 16, 2026)