Design deep neural network architecture using a genetic algorithm for estimation of pile bearing capacity

Area of Science:

• Geotechnical Engineering
• Artificial Intelligence in Civil Engineering
• Machine Learning Applications

Background:

• Accurate prediction of pile bearing capacity is crucial for safe and economical foundation design.
• Traditional methods can be time-consuming and may not capture complex soil-pile interactions effectively.
• Machine learning offers a promising avenue for enhancing predictive capabilities in geotechnical engineering.

Purpose of the Study:

• To develop and evaluate a hybrid Genetic Algorithm-Deep Learning Neural Network (GA-DLNN) model for predicting the bearing capacity of driven piles.
• To optimize the Deep Learning Neural Network (DLNN) model parameters using evolutionary algorithms.
• To assess the impact of feature selection on the predictive accuracy of the hybrid model.

Main Methods:

• A Genetic Algorithm (GA) was employed for feature selection from a dataset of 472 driven pile static load tests.
• A GA-DLNN hybrid model was developed to optimize DLNN parameters (network algorithm, activation function, hidden layers, neurons per layer).
• The dataset was split into training (60%), validation (20%), and testing (20%) sets for model development and evaluation using R2, IA, MAE, and RMSE metrics.

Main Results:

• The GA-DLNN hybrid model successfully identified optimal parameters for the DLNN, improving prediction accuracy.
• The model utilizing only the most critical features selected by the GA demonstrated superior accuracy compared to using all input variables.
• Performance metrics (R2, IA, MAE, RMSE) indicated the effectiveness of the hybrid model in predicting pile bearing capacity.

Conclusions:

• The GA-DLNN hybrid model provides an effective approach for optimizing DLNN performance in predicting driven pile bearing capacity.
• Feature selection using Genetic Algorithms significantly enhances the accuracy of the predictive model.
• This AI-driven methodology offers a robust and accurate alternative for pile foundation design analysis.