Abstract: To study the stress deformation law in the process of deep foundation pit excavation, a case study is conducted on a 34 mdeep foundation pit of Changchun metro, and the MIDAS GTS is adopted to simulate the foundation pit excavation process under various conditions, obtaining the surface settlement, deformation of retaining structure, and axial force variation law. Furthermore, the simulated results are compared with the monitoring results to verify the feasibility of the model. The surface settlements under various conditions are extracted according the model established, and the excavation depth, support amount, and dewatering volume are taken as input parameters to predict the surface settlement of the foundation pit. First, the maximum surface settlement extracted from the model is taken as the sample data, and the sample data are classified by Kmeans clustering algorithm to determine the basis function center for the radial basis neural network. Then, the Kmeans clustering algorithm Kradial basis function(RBF) prediction model is constructed to predict the surface settlement. The applicable results show that the surface settlement prediction results by KRBF prediction model is closer to the actual monitoring data than that of traditional RBF model. Compared with the actual monitoring data, the mean square error, the mean absolute error, and the maximum absolute error of KRBF are 0.212 mm, 0.278 mm, and 1.11 mm, respectively, indicating a good prediction accuracy.

Key words: deep foundation pit, settlement prediction, clustering algorithm, neural network, numerical simulation, site monitoring