Probabilistic reliability assessment method for max ground settlement prediction of subway tunnel under uncertain construction information

Abstract

Ground settlement resulting from shield tunnelling in densely populated areas has a significant impact on the surrounding environment, while accurate prediction of max ground settlement (MGS) is challenging under uncertain construction conditions. This paper investigates the vine copula probabilistic dependence approach for MGS predictions with incomplete information. A Monte Carlo simulation framework is established to incorporates vine copula analysis for eight identified soil parameters. Finite element (FE) method was used to model construction tunnels with different parameters and determine the MGS induced by excavation. The modelling results were used to construct six MGS base learners, which were created using six machine learning models combined with hybrid particle swarm optimisation (PSO) and gravity search algorithms (GSA). The integrated learning model combined six distinct base learners to generate a meta-learner. Improved hybrid GSA and PSO leveraged the global search capabilities of PSO and the local search abilities of GSA to optimize the integrated learning model. The FE model and meta-model predictions of MGS were validated using twelve uncertain input parameters. The results suggested that the hybrid GSA and PSO enhanced the precision of regression in the integrated learning model, and the resulting meta-model improved the reliability of MGS predictions in situations with uncertain information.