Abstract: As the limit state function of slope can't be explicitly expressed,conventional methods for slope reliability analysis are disadvantageous for difficulties and cumbersome calculation. A method for slope reliability analysis is proposed by combing the finite difference method of FLAC^3D and the extreme learning machine (ELM). Samples of random variables are generated through uniform experimental design, and the safety factors of these random variables are calculated through the strength reduction method of FLAC^3D. The mapping relationship between safety factors and random variables are obtained to construct the response surface function through the powerful fitting ability of ELM. Furthermore, a large number of random numbers generated by Monte-Carlo method are introduced into the function fitted by ELM to calculate the failure probability and reliability index of slope. Comparison with other methods through case study manifests that the proposed method is easy to be realized with reliable result.The research result provides a new approach for reliability analysis of slope, which is of broad application prospect.

Key words: slope reliability, extreme learning machine (ELM), response surface function, strength reduction method, Monte-Carlo simulation, failure probability