ABSTRACT
Long-term longitudinal settlement and horizontal convergence of metro tunnels are critical problems. To be able to forecast tunnel settlement and convergence for a distant future is crucial to pursue remedial measures in a timely manner. Existing research mostly focused on forecasting overall tunnel health condition combining multiple tunnel responses besides settlement and convergence, rather than forecasting them directly. Moreover, the forecasts are based on immediate earlier measurements of these responses and are unable to predict them into the future. This research aims to develop methods that are capable of forecasting tunnel settlement and convergence for future tunnelling operation based on early measurements. Machine Learning (ML) algorithms are trained to develop predictive models. The use and validity of the ML algorithms are demonstrated using data collected from the Metro Line 1 in Shanghai during its operational years. It is shown that the developed models could successfully predict tunnel longitudinal settlements and horizontal convergences as far as 14 years apart.
