Scholarlink

<p>Flashover is a sudden fire propagation that occurs within a room, where all items in the room bursting into the fire, making it one of the main causes of casualties. This paper presents the development of two models, the Ensemble of Long Short-Term Memory (E-LSTM) and the Ensemble of Gated Recurrent Unit (E-GRU), for predicting the occurrence of flashover in various compartment structures, and the development of Vision Transformer (ViT) to calculate the heat release rates in fire images supports the practical application of E-LSTM and E-GRU. Synthetic data from 1500 fire cases were collected, including temperature, heat release rates, oxygen volumetric fractions, carbon dioxide volumetric fractions, compartment floor area, and vent area, covering a wide range of fire scene conditions. ViT was trained on 4860 fire images, R<sup>2</sup> value of 0.9117 demonstrates the model accurately acquires the heat release rate in fire. E-LSTM and E-GRU, comprising 11 LSTM and GRU sub-models, achieved average accuracies of 94.88% and 95.76% respectively. In real fire scenario tests, E-LSTM and E-GRU exhibited accuracies of 88.31% and 93.90%, showcasing their ability to predict flashover occurrences with a high degree of precision within a 60&#xa0;s lead time. The results of this study indicate that the proposed machine learning models E-LSTM, E-GRU, and ViT can provide support for smart firefighting, reducing casualties and property losses.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine Learning Based Flashover Prediction Models Using Synthetic Data and Fire Images

  • 期刊论文

  • 发表日期：

    2025年2月11日

查看全文
  • Yansheng Song,
  • Guang Xiao,
  • Haoran Wang

摘要

Flashover is a sudden fire propagation that occurs within a room, where all items in the room bursting into the fire, making it one of the main causes of casualties. This paper presents the development of two models, the Ensemble of Long Short-Term Memory (E-LSTM) and the Ensemble of Gated Recurrent Unit (E-GRU), for predicting the occurrence of flashover in various compartment structures, and the development of Vision Transformer (ViT) to calculate the heat release rates in fire images supports the practical application of E-LSTM and E-GRU. Synthetic data from 1500 fire cases were collected, including temperature, heat release rates, oxygen volumetric fractions, carbon dioxide volumetric fractions, compartment floor area, and vent area, covering a wide range of fire scene conditions. ViT was trained on 4860 fire images, R2 value of 0.9117 demonstrates the model accurately acquires the heat release rate in fire. E-LSTM and E-GRU, comprising 11 LSTM and GRU sub-models, achieved average accuracies of 94.88% and 95.76% respectively. In real fire scenario tests, E-LSTM and E-GRU exhibited accuracies of 88.31% and 93.90%, showcasing their ability to predict flashover occurrences with a high degree of precision within a 60 s lead time. The results of this study indicate that the proposed machine learning models E-LSTM, E-GRU, and ViT can provide support for smart firefighting, reducing casualties and property losses.

Graphical Abstract