A multiple back propagation neural network fusion algorithm for ceiling temperature prediction in tunnel fires. (1st April 2023)

• Record Type: Journal Article
• Authors: Sun, Bin; Xu, Zhao-Dong; Zhou, Haijun
• Is Part Of: Engineering structures. Volume 280(2023)
• Publication Date: 2023-04-01
• Subjects: Back propagation neural network -- Ceiling temperature distribution -- Tunnel fires -- Different ventilation conditions -- Large-scale tunnel pool fire experiments; Structural engineering -- Periodicals; Structural analysis (Engineering) -- Periodicals; Construction, Technique de la -- Périodiques; Génie parasismique -- Périodiques; Pression du vent -- Périodiques; Earthquake engineering; Structural engineering; Wind-pressure; Periodicals; 624.105
• DOI: 10.1016/j.engstruct.2023.115601
• Languages: English
• ISSNs: 0141-0296

Abstract

Highlights: A multiple neural network fusion algorithm is developed to study tunnel fires. Ceiling temperature distribution in tunnel fires can be predicted with a high precision. Many sub-networks are constructed to a multiple neural network in the algorithm. Large-scale tunnel pool fire experiments were conducted to verify the algorithm. Abstract: This study presents a multiple back propagation neural network (BPNN) fusion algorithm to predict the ceiling temperature distribution and the maximum ceiling temperature in tunnel fires for better firefighting decision-making. The algorithm belongs to a kind of data-driven artificial intelligence methods, which is not limited to the special tunnel fire scenes. To improve the prediction precision, a strong multiple BPNN including some sub-BPNNs and an additional external iteration procedure are used in the algorithm. Large-scale pool fire experiments under different ventilation conditions were carried out to investigate the ceiling temperature character of tunnel fires and verify the algorithm. The results show that the algorithm can predict well the ceiling temperature distribution and the maximum ceiling temperature under different ventilation conditions, which has a higher prediction precision than the traditional single BPNN algorithm and some developed theoretical models.