A generic graph neural network-based model is developed to predict the potential occurrence of flashover for different building structures. The proposed model transforms multivariate temperature data into graphstructure data. Utilizing graph convolution operations, the temporal dependencies and spatial correlations of the temperature data are captured. Model assessment show that the generic flashover prediction model can distinguish different building structures and provide forecasts in advance to classify the potential occurrence of flashover with an overall accuracy of~ 93%. This work constitutes a machine learning-based forecasting model framework accounting for a wide range of building structures. The research outcomes from this study are expected to facilitate data-driven fire fighting, leading to enhanced situational awareness and improved fire fighting safety to help reduce US fire fighter deaths and injuries.
|Publication status||Published - Dec 2021|
|Event||AOSFST 2021–12th Asia-Oceania Symposium on Fire Science and Technology - |
Duration: 7 Dec 2021 → 9 Dec 2021
|Competition||AOSFST 2021–12th Asia-Oceania Symposium on Fire Science and Technology|
|Period||7/12/21 → 9/12/21|