Study on temperature prediction and early warning of building resistive electrical circuit fire: A mathematics model

作者：Sheng He, Jinfu Guan, Jingwu Wang, Kun Wang, Xueming Shu, J.L. He, Jiale Zhang, Song Chen, Wenguo Weng · 发表于：Results in Engineering · 年份：2025 · DOI：10.1016/j.rineng.2025.108223 · 被引用次数：4 · 研究领域：Fire dynamics and safety research、Electrical Fault Detection and Protection、Fire effects on concrete materials

• Proposed a novel method for building electrical fire temperature prediction. • Physical process and mathematics derivation of prediction were solved. • Temperature step-change is linear-corrected via cumulative variety. • Temperature probability normal dist matches actual values well. Building electrical circuit temperature perception and prediction are crucial for preventing the building electrical fire risks. Nevertheless, existing engineering prediction methods often lack sufficient accuracy and timeliness by virtue of the fixed threshold fire warning or failure to predict the temperature step-change. In this study, a novel mathematics method was proposed to dynamically predict the building electrical circuit temperature. First, the Long Short-Term Memory (LSTM) neural network was selected as the basic predictive framework. Two types of three-phase electrical overload experiments (short term single-overloading and long term periodic-overloading) were implemented to establish the training dataset (temperature, voltage, current, and residual current) which were distinguished by sampling conditions and frequencies Second, a cross model was built to validate the prediction accuracy with conditions transformation. Third, the temperature step-change was calibrated via the linear relationship with accumulated temperature variation and the usage of complex structure LSTM. Subsequently, the second-order temperature residual as well as its normality test were calculated. Finally, ...