Prediction of surface settlement caused by tunneling with ARMA based time-series decomposition

Abstract

A reasonable prediction of surface settlement caused by tunnel excavation in complex environments is of great importance for ensuring the safety of surface structures. The introduction of machine learning (ML) and deep learning (DL) provides a new solution for surface settlement prediction. A novel optimisation method to enhance the performance of ML and DL is proposed. This method employs an autoregressive moving average (ARMA)-based Time-series Decomposition, abbreviated as ATD model. First, the surface settlement time-series was decomposed into two subseries features, construction-caused and stochastic-caused surface settlement, using the ATD model. Second, four ML algorithms and five DL algorithms based on unsupervised learning were introduced to predict the two subsequences. Then, the prediction results of the sub-sequences were linearly combined to obtain the predicted values of surface subsidence. The efficacy of each ML and DL algorithm was markedly enhanced following the implementation of the ATD model outlined in this study. The average 1-R² decreased by 32.62 %, whereas the Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE) exhibited mean reductions of 19.91 %, 23.37 %, and 16.44 %, respectively. The critical error value was introduced for the analysis of the frequency of the occurrence of model errors. The results demonstrate that the utilisation of the ATD model can significantly reduce the prediction error associated with ML and DL for surface settlement prediction.