Identification of Distributed Lateral Earth Pressure in Excavations Based on High-Dimensional Bayesian Inference

For excavations supported by servo steel struts, a reasonably accurate estimation of in situ lateral earth pressure on diaphragm walls is crucial for the dynamic design of active axial force adjustments in these struts. However, theoretical methods struggle to precisely estimate earth pressure due to the inability to fully capture the complex factors in the construction process, while direct monitoring is often impractical because of high costs. In this study, a Bayesian inference framework is proposed to identify lateral earth pressure based on measured wall deflections. Three-dimensional load–structure models are employed as forward models to account for the spatial distribution of earth pressure, which is discretized into grids and represented as a high-dimensional nodal force vector for precise inversion. Efficient algorithms are employed to sample the posterior distribution of unknown force parameters. The effectiveness of the framework is first validated through a numerical case, in which a 231-dimensional nodal force vector is identified. The impacts of deflection observation sparsity and measurement error on identification accuracy are also analyzed through the case. Furthermore, an experimental case is used to further evaluate the performance of the framework in more realistic scenarios. The overall distribution of the identified pressure agrees well with the true distribution, and the 95% confidence intervals successfully contain the true values. The results demonstrate the applicability and effectiveness of the Bayesian inference framework, and some suggestions are drawn from the results.