Abstract
The stability of slopes is an important parameter which can affect many engineering projects. In this study, we employed genetic programming (GP) and artificial neural network (ANN) techniques, based on upper bound (UB) limit analysis, for the problem in designing solution charts for slope stability. Existing theories of genetic programming predictive network models have not been applied in the area of slope stability. Accordingly, the main objective of this research is to propose a new GP model to estimate the factor of safety parameter and providing design solution charts in a two-layered cohesive slope. A dataset containing 400 UB analysis models was used to train and test the GP and ANN networks. Variables of the GP algorithm training network parameters and weights such as population size, number of genes, and tournament size were optimized. The input includes d/H, (depth factor), the undrained shear strength ratio (Cu1/Cu2), and slope angle (β), where the output was taken as a dimensionless stability number (N2c). The predicted results for both datasets (training and testing) from the GP and ANN models were evaluated based on two statistical indexes (root mean square error, RMSE, and coefficient of determination, R2). Besides, the obtained results were compared with actual values of N2c, in the form of design charts. The results show that both the GP and ANN models are accurate enough to be used in this field. Also, ANN performed slightly better than the GP. As a result, a formula was derived for each GP and ANN models to assess the slope stability behaviors of two-layered cohesive soils.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call