Swarm-based analysis through social behavior of grey wolf optimization and genetic programming to predict friction capacity of driven piles

• Published in: Engineering with computers 2021-04, Vol.37 (2), p.1277-1293
• Main Authors: Moayedi, Hossein, Mu’azu, Mohammed Abdullahi, Kok Foong, Loke
• Format: Article
• Language: English
• Subjects: CAE) and Design; Calculus of Variations and Optimal Control; Optimization; Classical Mechanics; Computer Science; Computer-Aided Engineering (CAD; Control; Data mining; Datasets; Driven piles; Genetic algorithms; Math. Applications in Chemistry; Mathematical and Computational Engineering; Neural networks; Optimization; Original Article; Prediction models; Sensitivity analysis; Shear strength; Structural analysis; Systems Theory; Training
• ISSN: 0177-0667; 1435-5663
• DOI: 10.1007/s00366-019-00885-z

The advantage of new data mining-based solutions, and more recently, optimization algorithms (i.e., basically swarm-based solutions) have enhanced traditional models of engineering structural analysis. This paper investigates social behavior of Grey Wolf Optimization (GWO) in improving the neural assessment of friction capacity ( f s ) of concrete driven pile systems. Besides, the genetic programming (GP) algorithm was also proposed to have comparison with the proposed GWO prediction outputs. To achieve this goal, four f s influential factors of pile length (m), pile diameter (cm), effective vertical stress ( S v ), and undrained shear strength ( S u ) are considered for preparing the required dataset. A swarm size-based sensitivity analysis is then carried out to use the best-fitted structures (i.e., more convergency in the final output) of each ensemble. The results of the best prediction network from both above-mentioned sensitivity analyses were compared. The results show that both GWO and GP models presented excellent performance. The findings of neural networks varied based on the number of neurons in a single hidden layer and of course the level of its complexity. Based on R 2 and RMSE, values of (0.9537 and 9.372) and (0.8963 and 7.045) are determined, for the training and testing datasets of MLP-based solution, respectively. On the contrary, for the GP and GWO-MLP proposed predictive models, the R 2 of (0.9783 and 0.982) and (0.913 and 0.892) were found for the training and testing datasets. This proves the better performance of GWO when combined with MLP in predicting engineering solutions comparing to conventional MLP or GP-based combinations.