Abstract: The complexity of slope engineering is not only reflected in the variability of geotechnical parameters, but also in the implicit and nonanalytic properties of the performance function. Therefore, the direct calculation of slope stability reliability under implicit performance function that based on Janbu method is researched. First, the slope limit equilibrium model is called to obtain basic geotechnical parameters, and then the initial samples that affect slope stability is acquired through the Latin hypercube sampling. Secondly, using the Kriging anisotropic association mapping method dependence mapping method, the slope function value is expressed as a function of the random process. Then combined with the active learning method, and based on the searching rules to adjust the training samples, the slope performance functions denoted by the random process that meet the accuracy of the actual project is determined through iterative cycle. Finally, the random process function is called to work out the failure probability of slopes through the checking point method (JC method). The case study shows that the accuracy of the proposed method is equivalent to that of the Monte Carlo simulation; but the calculation process is simple, highly efficient, and more practicable.

Key words: slope engineering, Kriging model, Latin hypercube design, most failure zone, active learning