Prediction of Active Earth Pressure in Constrained Backfill Retaining Walls using Support Vector Regression and Traditional Datafit based Non-Linear Regression

Constrained Backfill Retaining Walls (CBRWs) are increasingly utilized in geotechnical engineering, particularly in mountainous regions and densely populated urban areas where space limitations demand innovative design solutions.These walls exhibit unique geometric configurations, leading to complex earth pressure behaviors that deviate significantly from those observed in conventional retaining walls.Accurate prediction of the coefficient of active earth pressure (Ka) is crucial for ensuring the stability and safety of CBRWs.However, traditional methods using data-fitting software, though often yielding high R² values (up to 0.99), may not provide reliable predictions in real-world applications.To address these shortcomings, this study investigates the application of machine learning (ML) techniques, specifically Support Vector Regression (SVR), as an advanced method for predicting Ka in CBRWs.The current investigation models three failure mechanisms involving single, double, and triple rigid blocks, depending on the distance between the rock face and the retaining wall, to generate a dataset of Ka, calculated through in-house MATLAB codes.These computations are both time-consuming and computationally intensive, highlighting the necessity for an efficient predictive model.The SVR model is employed to predict Ka, and its performance is compared against conventional data-fitting approaches.The input parameters for the predictions include -aspect ratio (b/h), internal frictional angle of the soil ( φ ), rock-face angle (η ), retaining wall inclination ( β ), backfill inclination ( ε ), interface friction angle between rock-face and soil ( δ ), and interface friction angle between retaining wall and soil (ψ ).The model effectiveness is evaluated using multiple error metrics.The SVR model achieved a coefficient of determination (R²) value of 97.4%, demonstrating strong predictive capability.In contrast, conventional data-fitting software exhibits limitations in accurately capturing the complex Ka behavior in CBRWs.The findings of this study underscore the potential of SVR to enhance the design and analysis of CBRWs by offering greater accuracy and computational efficiency compared to traditional methods.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.