Optimizing the architecture of the artificial neural network by genetic algorithm to improve the predictability of pile bearing capacity based on CPT results

Abstract

This paper presents the results of applying the Artificial Neural Network (ANN) model in determining pile bearing capacity. The traditional methods used to calculate the bearing capacity of piles still have many disadvantages that need to be overcome such as high cost, complicated calculation, time-consuming. Currently, Artificial Intelligence (AI) is a useful tool that is applied in many fields to save time and costs. The study develops an ANN model and optimizes the architecture, using the Genetic Algorithm (GA) to determine the pile bearing capacity. A dataset of 108 pile static compression results is used to train and test the model. The results of the study are compared with the experimental formula according to Vietnamese nation standard TCVN 10304:2014, showing that the ANN model with well optimized, allowing prediction of pile bearing capacity close to experimental results and better than the formula in nation standard. Specifically, the ANN model gives 12% and 32.4% better performance, respectively, than the empirical formula on R2 and RMSE criteria, respectively. The results of the study are a premise for the application of AI in solving pile problems in the field of construction.