An efficient optimal neural network based on gravitational search algorithm in predicting the deformation of geogrid-reinforced soil structures

The deformation of a Geosynthetic reinforced soil (GRS) structure is a key factor in designing this type of retaining structures. On the other hand, the feasibility of artificial intelligence techniques in solving geotechnical engineering problems is underlined in literature. This paper is aimed to show the workability of two soft computing techniques in predicting the deformation of GRS structures. For this reason, first a relevant case study was modelled into ABAQUS, a finite element (FE) software. Then, the FE results (GRS deformations) were checked against the recorded deformations of the full-scale test. Subsequently, 166 finite element analyses were performed for dataset construction. Then, two predictive models of GRS deformations were constructed. For intelligent model construction, two artificial neural networks (ANN) were coupled with Gravitational Search Algorithm (GSA) and Particle Swarm Optimization (PSO), respectively. It was found that both GSA-based ANN and PSO-based ANN predictive models work good enough. However, the correlation coefficient (R) of 0.981 as well as the system error of 0.0101 for testing data suggest that the GSA-based ANN predictive model outperforms the PSO-based ANN model with R value of 0.973 and system error of 0.0127. Overall, findings recommend that the proposed models can be implemented in assessing the performance of geosynthetic reinforced soil structures.

Automated summary

This study investigates the feasibility of two soft computing techniques in predicting the deformation of Geosynthetic reinforced soil (GRS) structures, showing that the Gravitational Search Algorithm (GSA)-based artificial neural network (ANN) model outperforms the Particle Swarm Optimization (PSO)-based ANN model. The findings suggest that the proposed models can be implemented in assessing the performance of geosynthetic reinforced soil structures.