A Hybrid Optimization Approach for Pile Capacity Estimation Using Radial Basis Functions and Time-Dependent Variables

Abstract

Since the pile and surrounding soil interact in complex, nonlinear ways and time-dependent geotechnical parameters play a significant role, accurately estimating the axial bearing capacity of driven piles is still a crucial but difficult part of geotechnical engineering. The new hybrid predictive framework proposed in this study combines two recent metaheuristic optimization algorithms, Leader Harris Hawks Optimization (LHHO) and Graylag Goose Optimization (GGO), with the Radial Basis Function (RBF) model. Through the incorporation of temporal variations, such as changes in soil resistance due to pore water pressure dissipation, that are commonly overlooked in conventional modeling approaches, the main goal is to improve the predictive accuracy and generalization capability of pile capacity models. By combining time-dependent variables with a correlation-based feature selection mechanism, this work is novel in that it enables the identification of the most important input parameters and the removal of redundant or unnecessary features. In comparison to actual pile load test data, the resulting hybrid models, known as RBGG and RBLH, perform better. During training, the RBLH model demonstrated high accuracy and reliability with an R2 value of 0.986 and an RMSE of 183.680. In addition, sensitivity analysis showed that the most important factors affecting pile capacity are unit weight, soil cohesiveness, and pile length. With major ramifications for enhancing safety, decreasing uncertainty in construction procedures, and optimizing foundation design, the suggested methodology provides engineers with a strong, data-driven tool.