Predicting friction capacity of driven piles using new combinations of neural networks and metaheuristic optimization algorithms

Abstract

Friction capacity is a principal characteristic in designing driven piles. Considering the complexities in analyzing the behavior of piles, many studies have recommended the use of machine learning for this purpose. However, the used methodologies need to be updated and improved with respect to recent computational advances such as the development of optimization algorithms. In this work, three metaheuristic algorithms, namely equilibrium optimizer (EO), biogeography-based optimization (BBO), and salp swarm algorithm (SSA) are deployed to optimize an artificial neural network (ANN) for predicting pile friction capacity based on pile geometry, effective stress, and shear strength. The findings indicate the suitability of the proposed algorithms. More specifically, in the training phase, the ANN supervised by SSA yielded the most accurate results, whereas in the testing phase, the BBO-ANN outperformed the two other models. The calculated mean absolute error, Pearson correlation coefficient, and root mean square error for the models are as follows: 6.0740, 0.9385, and 7.0678 for the EO-ANN, 6.1450, 0.9440, and 6.7343 for the BBO-ANN, and 5.9684, 0.9395, and 7.1322 for the SSA-ANN. It is shown that both SSA-ANN and BBO-ANN can serve as efficient tools for the reliable design of driven piles, providing efficient computational intelligence alternatives to traditional design methods.