Neural Network-Particle Swarm Optimization Model for Predicting Slope Stability of Homogeneous Earth Dams

Slope stability of homogenous earth dams is a very important design analysis to consider since earth dams are prone to failure on slopes due some factors such as material properties and the design slope of the downstream side. SLOPE/W simulation using the method of slices was utilized to determine the effect of the input parameters in a Mohr-Coulomb analysis including the unit weight, cohesion, and angle of internal friction (Ø) of the dam material to be used. Hydraulic structure designers encounter difficulties in analyzing the best design which include the proper slope due to the complexity of the required input parameters and other factors affecting the stability of the dam. It is the main objective of this paper to provide a prediction model of the slope stability (SS) of homogenous earth dams using neural network (NN) particle swarm optimization (PSO) model. The model output will be of great contribution to the designers in investigating the proper earth material and slope design to consider both the structural integrity and economic aspect of the structure. There were seven (7) slopes and 14 earth materials analyzed in the study. The particle swarm optimization technique was implemented and utilized to train and optimize the initial neural network model. The material parameters such as unit weight of the soil (kN/m
³
), cohesion (kPa), angle of internal friction Ø (degrees), and slope were utilized as the input variables in the study while the factor of safety (FOS) values was used as the output variable. The developed NN-PSO model has an R (all) value of 0.99752, while the MSE of the developed NN-PSO model is 0.0021844. The values of R and MSE are close to the ideal values of 1.0 and 0 respectively, showing further that the model is satisfactory. Using the Garson’s algorithm, it was observed that the angle of friction is the most significant parameter to the FOS.