Research on Prediction of Shield Misalignment Based on LRTC-TNN and LSTM

This study introduces an innovative predictive model, LW-LSTM, combining Low-Rank Tensor Completion with Truncated Nuclear Norm (LRTC-TNN), Wavelet Transform (WT), and Long Short-Term Memory Neural Network (LSTM) to preemptively address shield misalignment in tunnel construction. The model begins with LRTC-TNN to fill in the missing shield time series data, enhancing dataset completeness. Wavelet Transform is then applied for data denoising, improving data accuracy for predictive analysis. The model's performance is demonstrated in tunnel projects, forecasting shield misalignment effectively. Extensive experiments show that LW-LSTM outperforms in predicting shield panel and elevation deviations, highlighting its potential in facilitating proactive adjustments in construction parameters, thereby reducing misalignment risks and enhancing overall project efficiency and risk management.