Efficient prediction of unsteady piston winds in subway tunnels based on machine learning

In emergencies such as fires in subway tunnels, the rapid prediction of piston winds therein is needed to provide theoretical guidance for emergency decision-making and fire rescue. Previous research has neglected the efficiency of piston wind prediction. For this reason, to improve on traditional methods, this paper uses eight different machine learning algorithms, including support vector regression (SVR), k-nearest neighbors (KNN), and extreme gradient boosting (XGBoost) et al., to construct a fast prediction model for unsteady piston winds in subway tunnels. A relatively optimal machine learning algorithm was determined through various evaluation methods, and feature importance analysis was carried out. The results show that it is feasible to utilize a machine-learning method for the fast prediction of unsteady piston winds in subway tunnels. The RF algorithm had a high prediction accuracy for unsteady piston winds, and the Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Symmetric Mean Absolute Percentage Error (SMAPE), and Coefficient of Determination (R²) of the prediction results were 0.04, 0.05, 1.15 %, and 0.995, respectively. The influence degree of each factor on the subway tunnel piston wind, in descending order, is as follows: time t>blockage ratio β>train running speed u>tunnel length l_tu>train length l_tr.