Machine learning prediction and analysis of supplement velocity and maximum smoke temperature rise in half-U-shaped tunnel fire
The coupling of the fire temperature field and wind field in a half-U-shaped tunnel makes it difficult to quantify the temperature change in the tunnel,thus the theoretical derivation is difficult,the hypothesis is oversimplified,and the prediction model is relatively complicated.Therefore,this paper uses FDS numerical simulation and machine learning to conduct simulation analysis and machine learning prediction of 320 groups of fire conditions.The data set of fire-related parameters of a half-U-shaped tunnel is composed of the numerical simulation of fire with different working conditions.The simulation results and the predictions of different machine learning models for air refill rates and maximum smoke temperature rise are analyzed,The results show that the increase of slope height and heat release rate can improve the air replenishment velocity,while the increase of the longitudinal wind velocity reduces the supplement velocity by restraining the rise of the fire temperature field.The prediction effect of the BP neural network on the test set and training set is more accurate than other machine learning models.The coefficient of determination R2 can reach 0.99,and the average absolute error can be as low as 0.058.According to the Shap value,the characteristic factors affecting the air replenishment velocity in the tunnel are ranked in order of importance as height effect,thermal effect,and wind effect.The maximum smoke temperature rise is affected by the velocity.When the slope height is small,the temperature rise sharply decreases with the velocity,while the velocity has no significant effect on the maximum smoke temperature rise when the slope is high.This phenomenon accords with the result of theoretical analysis.Compared with other machine learning methods,both the BP neural network and theoretical calculation can accurately predict the maximum temperature rise of smoke,and R2 is greater than 0.9.
万方数据
维普
