Spatio-Temporal Prediction Surface Displacement in Urban Underground Excavation: A Case Study in Seville"

Abstract

One of the primary challenges in excavating underground in urban areas is controlling and mitigating ground surface displacement caused by Earth Pressure Balance (EPB) tunneling. It is crucial to avoid damaging historical monuments and buildings in these areas. This paper presents a new method for predicting the surface displacement caused by EPB in Seville. A spatiotemporal dataset was generated for this study using numerical simulation in FLAC3D. The simulation replicates the excavation process of the Seville metro line in real-time, and records the surface displacements at selected points in the dataset. The last 20-time steps of excavation are predicted, and the first 80-time steps are chosen for training and tuning hyperparameters, as the dataset is spatiotemporal. A recurrent neural network (RNN) is used to detect and predict patterns between surface displacement and input features at different time steps and locations of the excavation. After fine-tuning the RNN, the model achieved an accuracy of 0.91 for the evaluated R-squared (R2), indicating its practicality for real-time prediction of surface displacement in underground excavations in Seville. The model's performance can be further improved with a larger data range. By deploying it as a hazard detector, the model can issue a warning if the ground displacement exceeds the limit, thereby preventing potential hazards. This approach can help control and mitigate potential hazards in underground excavations in historical cities.

Full Paper