The Application of Support Vector Machine (SVM) Regression Method in Tunnel Fires

Abstract

Abstract Due to the relatively closed and narrow space of tunnel, when the fire happens, smoke spreads rapidly and evacuation is difficult. Because the conditions are changing of smoke and temperature distributions after fire burning with different situations, such as ventilation mode and fire source character and so on, it is important to study all the related issues with various fire conditions and find the optimal fire proof design method. This paper takes a fire prevention zone of a real highway tunnel as the reduced scale model (1:9) of tunnel fire test platform. The orthogonal tests of fire in tunnel model under different situations are carried out, with different fire heat release rates and ventilation velocities and fire source locations. The support vector machine (SVM) regression method is used to analyze the CO concentration and temperature in two references firstly, which shows the availability for tunnel fires in steady stage. Then CO2 gas concentration and movement are analyzed during the fire initial burning state based on the SVM regression method for our experiments. It is revealed that CO2 gas concentration in different situations can be predicted during the fire burning initial stage, and the orthogonal tests accompanied with the SVM regression can indicate the influence of various fire conditions for concentration and further guide the experiment scheme.