Probabilistic Back Analysis and Deformation Prediction by Integrating Multi-source Monitoring Data from Foundation Pits

To address uncertainties in soil parameters, prediction models of excavation response, and field monitoring data, this study proposes a probabilistic back analysis and deformation prediction method that incorporates multi-source monitoring data from foundation pits. Based on two types of deformation monitoring data, three distinct updating scenarios are established. The effectiveness of the proposed method is verified through a case study of a deep excavation, and the differential impacts of using single-source versus multi-source monitoring data on probabilistic back analysis are systematically investigated. The findings indicate that updated model parameters provide a more reliable basis for predicting the responses of retaining structures and soil in subsequent excavation stages, with the relative prediction error reduced from 36.6% to 0.18%. The single-source updating scenarios only significantly reduce the uncertainties in one type of model error and the corresponding response predictions, with the coefficients of variation (COV) decreased by up to approximately 76.5% and 96.7%, respectively. In contrast, the multi-source updating scenario simultaneously and significantly mitigate uncertainties in two types of model errors and their associated predictions, with COV reduced by up to approximately 88.0% and 96.9%, respectively. Compared to methods that update only soil parameters, the proposed approach, which updates both soil parameters and model errors, produces deformation predictions with greater accuracy and lower variability, thereby offering more reliable results. These findings can provide methodological guidance for enhancing predictions of excavation response through monitoring data.