Use of artificial neural networks for the prediction of time-dependent air speed variation in metro stations

Abstract

In this study, the time-dependent, induced air speeds at critical sections of underground metro stations are assessed using a novel one-dimensional data-driven approach. For this purpose, three artificial neural networks are used, each trained for the most basic configuration of a single train moving in a single tunnel. The first two are trained to provide the maximum and time averaged values of the induced air speeds while the train is moving inside the approach tunnel or the station. The third one is used to simulate the time-dependent air speed variation during train stoppage and departure. Typical structures of a metro system such as staircases and ventilation shafts are introduced into the solution using simple analytical relations based on loss coefficients. The developed approach is tested using two different metro stations that are currently in operation in Turkey. The selected stations are constructed using different tunneling techniques resulting in different air flow characteristics. The results show that the time variation of the air speed predicted by the developed model is, in general, in good agreement with the results of the Subway Environmental Simulation software, although further studies are necessary to model the acceleration and deceleration of trains more realistically.