Predicting the CPT-based pile set-up parameters using HHO-RF and WOA-RF hybrid models

• Published in: Arabian journal of geosciences Vol. 15; no. 7
• Main Authors: Duan, Lijuan, Wu, Miao, Wang, Qiong
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.04.2022; Springer Nature B.V
• Subjects: Algorithms; Cone penetration tests; Dimensions; Earth and Environmental Science; Earth science; Earth Sciences; Mathematical models; Neural networks; Optimization; Original Paper; Overburden; Parameters; Performance prediction; Piles; Prediction models; Root-mean-square errors; Skin friction; Skin resistance; Soil types; Optimization algorithms; Cone penetration test; Random forest; Pile set-up parameter; Prediction
• ISSN: 1866-7511; 1866-7538; 1866-7538
• DOI: 10.1007/s12517-022-09843-4

During the estimation of pile strength ( R t ), a parameter is utilized called pile set-up parameter ( A ). The value of this parameter was so controversial because it relates to many factors, such as the material of pile, the resistance of pile, type of pile, dimension of pile, and type of soil. So, developing prediction models based on neural networks could solve this matter. Therefore, this study aimed to develop hybrid random forest (RF) models to predict the pile set-up parameter ( A ) from cone penetration test (CPT) for the design aim of the projects. To this goal, the essential hyperparameters of the RF model were tuned by applying the whale optimization algorithm (WOA) and Harris hawks optimization (HHO). The selected variables as input were average corrected cone tip resistance, average skin friction, and average overburden pressure. Results show that both models have acceptable performance in predicting the set-up parameter A , with coefficient of determination ( R 2 ) larger than 0.8699, representing the admissible correlation between observed and predicted values. It is understandable that, in both the learning and validating phase, HHO-RF has better proficiency than the WOA-RF model, with R 2 and root mean square error (RMSE) equal to 0.9436 and 0.0307 for the training phase, and 0.8866 and 0.0371 for testing data, respectively. Moreover, by considering all performance accuracy criteria, the results demonstrate the ability of the HHO algorithm in determining the optimal value of RF hyperparameters than WOA.