关键词: 土质边坡;位移预测;失稳概率;概率反分析;代理模型;

Abstract: The traditional back analysis of slope displacement adopts the deterministic method, which ignores the non-uniqueness of parameters. In contrast, the probabilistic back analysis method can determine multiple sets of input parameters by quantifying the uncertainty of input parameters, which is of great significance for prediction of slope instability probability. In this paper, the uncertainties of input parameters were quantified by prior distribution, and then a high efficient surrogate model for the original slope numerical model was constructed. The input parameters were updated by Bayesian inference and Markov chain-Monte Carlo method, which greatly reduced the uncertainty of the input parameters. Then, by predicting the displacement of a slope and comparing with the actual monitoring displacement, the reliability and accuracy of the proposed method were verified. Finally, the slope instability probability was predicted by constructing an efficient active learning agent model. The prediction results can provide important support for quantitative risk assessment of slope’s instability.

Key words: soil slope; probabilistic back analysis; agent model; displacement prediction; instability probability