Prediction of the limiting uniform normal pressure in deep contiguous piled walls using soft computing techniques

Abstract

In urban civil engineering projects, deep contiguous piled walls are crucial for support and stability, particularly in limited spaces. This study employs advanced soft computing techniques, integrating FELA simulations, XGBoost, and optimization algorithms (PSO, GWO, and WOA), to predict the limiting uniform normal pressure behind circular and I-shaped contiguous piled walls in cohesive soil. The key factors in the prediction model include the soil properties, pile wall geometry, and soil strength parameters such as the spacing-to-width ratio (S/B or S/D), adhesion factor (m), height-to-width ratio of the I-section (H/B), and friction angle (φ). The sensitivity analysis reveals that for circular-shaped piles, φ is the most influential parameter (R = 0.69, R_i = 47 %), followed by the m and S/D, with moderate impacts (R = 0.49 and 0.28, R_i = 34 % and 19 %, respectively). For I-shaped piles, the m has the highest effect (R = 0.52, R_i = 28 %) followed by the S/B (R = 0.49, R_i = 26 %), H/B (R = 0.45, R_i = 24 %) and φ (R = 0.41, R_i = 22 %). The predictive performance of the constructed model was assessed using several performance metrics, regression plot, residual error histogram and regression error characteristic (REC) curve. The hybrid XGBoost-WOA model is identified as the most effective for both circular and I-shaped piles based on various performance metrics (R², RMSE, MAE, WMAPE, VAF, and PI) and error analyses. This approach aims to improve prediction accuracy and address the limitations of traditional methods in deep contiguous piled wall design.