Research on the deformation prediction of deep foundation pit support structures based on SSBA model

Abstract

[Objective] The deformation of support structures is crucial for the stability of deep excavations. To more accurately predict the deformation of support structures caused by deep excavation, [Methods] this paper proposes a novel deformation prediction model, SSBA, which integrates Spearman correlation coefficient, Spatio-temporal Convolutional Neural Networks(STCNN), Bi-directional Long-Short Term Memory, and Attention mechanism. [Results] Experimental result reveal that the deformation of the pit is positively correlated with excavation depth and phase, and negatively correlated with the support strut axial force. Excavation depth having the most significant impact on deformation. Analysis of the monitoring data revealed that the deformation of the support structure is less than the specified values, indicating that the internal support system effectively restricts the lateral displacement of the wall, and the support structure design is reasonable. Compared with four baseline models, the SSBA model achieved the smallest MAE and RMSE values and the largest R² value, indicating that it can predict the support structure deformation more accurately. The SSBA model can also predict the deformation values at different measuring points accurately, demonstrating good generalization capability and reliability. Through experiments conducted using field monitoring data from a certain foundation pit on Suzhou Metro Line 6, it was found that the SSBA model can more accurately predict diaphragm wall deformations, indicating that the model has good generalizability. [Conclusion] SSBA model can predict the deformation of support structures more accurately and provide guidance for the engineering construction.

Keywords

deep excavation / diaphragm wall deformation / parameter correlation / deep learning / attention mechanism