Research on Deformation Safety Risk Warning of Super-Large and Ultra-Deep Foundation Pits Based on Long Short-Term Memory
Yanhui Guo, Chengjin Li, Ming Yan, Rui Ma, Wei Bi
Abstract
This paper proposes transforming actual monitoring data into risk quantities and establishing a Long Short-Term Memory (LSTM) safety risk warning model for predicting the deformation of super-large and ultra-deep foundation pits in river–round gravel strata based on safety evaluation methods. Using this model, short-term deformation predictions at various monitoring points of the foundation pits are made and compared with monitoring data. The results from the LSTM safety risk warning model indicate an absolute error range between the predicted deformation values and on-site monitoring values of −0.24 to 0.16 mm, demonstrating the model’s accuracy in predicting pit deformation. Additionally, calculations reveal that both the overall risk level based on on-site monitoring data and the overall safety risk level based on predicted data are classified as level four. The acceptance criteria for the overall risk level of the foundation pit are defined as “unacceptable and requiring decision-making”, with the risk warning control scheme being “requiring decision-making, formulation of control, and warning measures”. These research findings offer valuable insights for predicting and warning about safety risks in similar foundation pit engineering projects.