BP neural network-based adaptive spatial-temporal data generation technology for predicting ceiling temperature in tunnel fire and full-scale experimental verification

This study proposes a novel adaptive spatial-temporal data generation method to achieve precise ceiling temperature perception in tunnel fires based on the modified BP neural network. In the method, an automatic update of training data set procedure is coupled in the BP neural network, which establishes an effective coarse-to-fine data generation method from coarse monitored data to fine ceiling temperature prediction in spatial-temporal scale. The method belongs to a kind of data-driven algorithms for ceiling temperature prediction in tunnel fires, which is not limited to the special fire scene. Additionally, full-scale fire experiment was conducted to verify the effectiveness of the method in China's largest tunnel fire experimental platform. Twenty-three thermocouples measured the ceiling temperature of the tunnel fire experiment. In contrast to the experimental results and the results obtained from the traditional BP neural network, the method can be used an effective numerical tool to predict the precise fine ceiling temperature spatial-temporal distribution, and the prediction precision is higher than the traditional BP neural network algorithm.