Obtaining detailed information on soil parameters is a significant precondition for slope rein- forcement and risk assessment. At present， the Markov chain Monte Carlo (MCMC) simulation is frequently used to update the statistical information of uncertain parameters based on in-situ and/or lab- oratory test data， but it is difficult to solve the problem of high-dimensional slopes due to a large amount of calculation consumption and poor convergence. In this paper， a slope displacement surro- gate model based on particle swarm optimization back propagation neural network is constructed to ac- celerate the calculation process. An improved Bayesian updating with subset simulation (BUS) is pro- posed for updating the statistics of soil parameters and slope reliability based on the monitoring data of slope displacement. The proposed method is then applied to a practical slope project (Changchun West Railway Station's Deep Foundation Pit Slope Project). The results indicate that the proposed method can effectively update the statistics of soil parameters， infer their posterior probability distribu- tion， and further update the probability of slope failure. Then， the slope displacement evaluated using the updated soil parameters agree well with the measured data， which confirms the applicability and ef- fectiveness of the proposed method. Additionally， after the Bayesian updating with the monitoring da- ta， the uncertainties of soil shear strength parameters are significantly reduced， but the probability of slope failure can be increased due to the influences of the ambient temperature， monitoring positions and values of monitoring data.
