A Reinforcement Learning approach to the location of the non-circular critical slip surface of slopes

Abstract

We propose a numerical procedure to locate the critical slip surface of slopes with the method of slices. We employ the Deep-Q Learning algorithm with experience replay and target memory to determine a non-circular slip surface. The overall stability analysis is performed with Janbu's simplified method. Our approach, however, is flexible and can accommodate other Limit Equilibrium Methods. The accuracy of the method is demonstrated by using typical verification examples and comparing the results to the search methods implemented in SLOPE/W and Slide2. It is demonstrated that the Deep-Q learning algorithm can be efficiently applied to layered slopes.

Topics & Concepts