Prediction of maximum ceiling temperature of rectangular fire against wall in longitudinally ventilation tunnels: Experimental analysis and machine learning modeling

Abstract

In this work, experimental analysis and machine learning modeling of maximum ceiling temperature for wall fire in longitudinally ventilation tunnels were conducted. The results indicate that the maximum ceiling temperature is restrained by the longitudinal ventilation, and the maximum ceiling temperature induced by a free fire is significantly lower than that of a wall fire. The aspect ratio of the burner has an impact on the maximum ceiling temperature. The larger the aspect ratio of the burner, the higher the maximum temperature. Based on the “mirror” assumption, the difference in maximum ceiling temperature caused by the changes in entrainment characteristics between wall fires and free fires can be explained. The influence of longitudinal wind inertia forces on the aspect ratio of the burner was analyzed. It was found that when the aspect ratio was large, it is affected by the smaller longitudinal wind inertia force to generate a higher flame height, and this will lead to higher maximum ceiling temperatures. The classical model cannot explain the difference in maximum ceiling temperature between wall fires and free fires, as well as the effect of aspect ratio on the maximum ceiling temperature. Further explanation and verification were conducted by combining the temperature field and the flow field. In addition, genetic algorithms were applied to compensate for the shortcomings of back propagation neural networks in randomly generating initial weights and initial biases. A new model-free calculation (GA-BPNN) model was proposed to predict the maximum ceiling temperature in a ventilation tunnel. Experimental results were compared with the predicted results, and error analysis was conducted on the data from the training and testing sets, with a relative error within ±20%. The GA-BPNN algorithm proved to be a promising method that can provide direct guidance for the experimental analysis and the determination of relevant parameters in the fire situation analysis.