Prediction Study of Ultimate Bearing Capacity of Super-Long Bored Piles Based on Random Forest and Data Augmentation

In this paper, we propose an innovative approach to solve the data scarcity problem and the bearing capacity prediction problem in the estimation of the ultimate bearing capacity of extra-long bored piles. To address the above problems, we first employ the variational autoencoder (VAE) technique to augment the limited drilled pile ultimate bearing capacity data, which significantly enlarges the size of the dataset and captures the potential distribution characteristics among the data at the same time. Subsequently, the augmented dataset is utilized to train a random forest model to achieve accurate regression prediction of the ultimate bearing capacity of extralong bored piles. In order to verify the effectiveness of the proposed method, we compared it with various machine learning algorithms. The experimental results show that the random forest model combined with VAE performs well on the task of estimating the ultimate bearing capacity of extra-long bored piles, which not only improves the training effect and generalization ability of the model, but also demonstrates significant advantages in a number of evaluation indexes. This study not only introduces an efficient data enhancement strategy for the civil engineering field, but also provides a reliable solution to solve the problem of predicting the ultimate bearing capacity of extra-long bored piles, which is expected to promote the theoretical development and practical application in this field.