TM-Informer-Based Prediction for Railway Ground Surface Settlement

With the acceleration of urbanization, subway construction has become a key part of urban development. However, shield construction often needs to cross existing railway lines, which requires accurate prediction of ground surface settlement of railway operation facilities to ensure railway safety and smooth construction. The existing Internet of Things (IOT) ground surface settlement prediction mainly relies on deep learning algorithms, but these algorithms are often difficult to take into account the extraction of local features and the capture of global features when processing long time sequence data, resulting in the prediction effect is affected when processing long-term dependent ground surface settlement data. In order to solve the above problems, this paper proposes a ground surface settlement prediction method based on T-BiGRU Mask Multi-Head Self-Attention Informer (TM-Informer). First, T-BiGRU is introduced into the model to effectively cover a wide range of time scales and extract the local feature of the settlement sequence data. Second, Mask Multi-Head Self-Attention (MMSA) is used to effectively solve the long-term dependence problem in long time sequence settlement data prediction and enhance the global feature capture ability of the model. In this paper, TM-Informer is compared with Informer, Transformer and LSTM by four evaluation indexes MSE, MAE, RMSE and MAPE. The experimental results show that the MSE, MAE, RMSE and MAPE of TM-Informer are 0.301, 0.432, 0.548 and 2.617, respectively, which are 0.08, 0.04, 0.06 and 0.26 higher than those of Informer, which proves the effectiveness of TM-Informer in railway ground surface settlement monitoring.