Real-time online prediction of surface settlement considering spatiotemporal characteristics during foundation excavation

The accurate prediction of surface settlement induced by foundation excavation is challenging owing to complex spatiotemporal characteristics. To address this challenge, a real-time online prediction model comprising input, offline, and online modules was proposed. Grey relation analysis was employed to extract effective spatial coupling variables to improve the quality of the data in the input module; a long short-term memory network was trained to capture temporal nonlinearity in the offline module; and a statistical process control cloud platform was embedded to realize online updating when predicting residual anomalies in the online module. The model was verified using a time series dataset from a metro foundation project and outperformed other existing models with high accuracy. Our results could help managers to control settlement in a timely manner to prevent disasters. Clarifying the spatial coupling scale to improve the quality of input data of the model is left for future work.