Abstract: This paper proposes a new method for modeling the probability distributions of shear strength parameters using maximum entropy principle (MEP). First, the procedure for constructing the probability density functions (PDFs) of shear strength parameters using MEP is presented. Then, Monte Carlo simulations are conducted to validate MEP for modeling the PDFs of shear strength parameters and probability of slope failure. Moreover, the accuracy and robustness of MEP, Akaike Information Criterion (AIC) and kernel density estimation (KDE) are compared. Finally, a dataset of shear strength parameters for residual soil is compiled, and an infinite slope is adopted to demonstrate the application of MEP to the estimation of the probability distributions of shear strength parameters and probability of slope failure. The results indicate that with limited shear strength data, MEP can effectively model the PDFs of shear strength parameters and probability of slope failure. In comparison with AIC and KDE, MEP produces more accurate and robust results for the PDFs of shear strength parameters and probability of slope failure. MEP not only avoids relying too much on the limited shear strength data in KDE, but also overcomes the possibility of excluding the true probability distribution in the set of candidate probability distributions in AIC. Furthermore, there are significant variations in the probability distributions of shear strength parameters and probability of slope failure estimated from limited data.

Key words: slope, reliability, shear strength parameters, probability distribution, maximum entropy principle