Consolidation Settlement Prediction Based on Seq2Seq-LSTM Considering Preloading

Structures constructed on soft ground can experience long-term settlement, which may lead to problems. Thus, settlement monitoring and management during construction are crucial for ensuring structural stability. Settlement predictions based on monitoring data are commonly performed using the hyperbolic, Hoshino, and Asaoka methods. However, these approaches estimate the final settlement based on the data obtained after completing preloading. This makes prediction challenging when the construction period is prolonged or when monitoring data are insufficient. To address these limitations, we propose a settlement prediction method based on a Seq2Seq-long short-term memory (LSTM) model that incorporates embankment load as an input factor. The model was trained on 38 datasets containing embankment load and settlement measurements from a specific site. In evaluations, the proposed model successfully predicted the settlement at different sites, demonstrating a relatively superior performance compared to the hyperbolic, Hoshino, and Asaoka methods. Therefore, the proposed approach can be useful for long-term settlement monitoring and management under soft ground conditions.